{"days":[
{"date":"2025-11-30","headline":"First edition of The Biosecurity Handbook published","entries":[{"category":"milestone","title":"First edition of The Biosecurity Handbook published","change":"Launches the handbook with Welcome, Preface, and References and How to Cite pages, the site structure, and a CC BY 4.0 license.","why":"Establishes the handbook's first public edition and how to cite it.","chapters":[["Welcome","/"],["Preface","/preface.html"],["References","/references.html"]],"sources":[]},{"category":"content","title":"Biosecurity foundations, emerging technology, AI risk, and governance","change":"The handbook runs to five parts and 20 chapters across foundations, classical biosecurity, emerging technology, AI and biosecurity, and governance, with 5 appendices. Red-Teaming AI Systems treats the ASL framework and uplift studies.","why":"Introduces the handbook and its treatment of AI safety levels and uplift studies.","chapters":[["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"],["What Is Biosecurity?","/foundations/what-is-biosecurity.html"],["The Biological Threat Landscape","/foundations/threat-landscape.html"],["Pathogens of Concern","/foundations/pathogens-of-concern.html"],["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"],["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"],["Gene Drives and Environmental Biosecurity","/biotechnology/gene-drives.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Building a Biosecurity Career: Federal Agencies, Fellowships, and AI-Bio Roles","/governance/biosecurity-careers.html"],["Case Studies in Biosecurity Failures and Successes","/governance/case-studies.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[]},{"category":"content","title":"What Is Biosecurity? defines the field and traces its history","change":"What Is Biosecurity? explains the biosecurity versus biosafety distinction and a historical timeline from 1925 bioweapons governance through the Biological Weapons Convention and the Soviet Biopreparat program.","why":"Explains the biosecurity and biosafety distinction and the field's historical context.","chapters":[["What Is Biosecurity?","/foundations/what-is-biosecurity.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Welcome","/"],["Preface","/preface.html"]],"sources":[]},{"category":"content","title":"Threat landscape and pathogens of concern chapters completed","change":"The Biological Threat Landscape classifies natural, accidental, and deliberate biological threats, covering zoonotic spillover (COVID-19, Ebola, pandemic influenza) and laboratory-acquired infections. Pathogens of Concern adds the CDC's Category A, B, and C threat agent classifications and the Federal Select Agent Program.","why":"Gives readers the handbook's core threat taxonomy before it turns to governance frameworks.","chapters":[["The Biological Threat Landscape","/foundations/threat-landscape.html"],["Pathogens of Concern","/foundations/pathogens-of-concern.html"]],"sources":[]}]},
{"date":"2025-12-01","headline":"Laboratory biosafety and biosecurity chapter completed; Dual-use research of concern chapter completed; Biological Weapons Convention chapter completed; Biosurveillance and early warning systems chapter added","entries":[{"category":"content","title":"Laboratory biosafety and biosecurity chapter completed","change":"Laboratory Biosafety and Biosecurity gets full content on the four Biosafety Levels (BSL-1 through BSL-4), primary and secondary containment barriers, and biosecurity measures such as personnel reliability programs, access controls, and select agent security requirements.","why":"Gives readers the handbook's core containment and security framework for laboratories handling dangerous pathogens.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[]},{"category":"content","title":"Dual-use research of concern chapter completed","change":"Dual-Use Research of Concern (DURC) defines the category and walks through the 2011-2012 H5N1 ferret-transmissibility controversy, the NSABB's reversal from recommending redaction to endorsing publication, and the resulting debate over redacting experimental details.","why":"Uses the H5N1 gain-of-function controversy as the handbook's central case study in dual-use publication oversight.","chapters":[["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"sources":[]},{"category":"content","title":"Biological Weapons Convention chapter completed","change":"International Governance and the Biological Weapons Convention covers the treaty's structure and core articles, the 2001 collapse of six years of verification protocol negotiations, voluntary confidence-building measures with roughly 50% participation, and the small, resource-constrained Implementation Support Unit.","why":"Explains why the treaty banning biological weapons has no inspection or enforcement mechanism.","chapters":[["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"]],"sources":[]},{"category":"content","title":"Biosurveillance and early warning systems chapter added","change":"A new chapter on Biosurveillance and Early Warning Systems covers the International Health Regulations (2005), indicator-based versus event-based surveillance, and major systems including GISRS, ProMED-mail, GPHIN, GISAID, and wastewater monitoring.","why":"Surveys how outbreak detection systems could also surface a deliberate biological release.","chapters":[],"sources":[]}]},
{"date":"2025-12-03","headline":"Medical countermeasures and biodefense chapter completed; Synthetic biology chapter added; countermeasures chapter expanded","entries":[{"category":"content","title":"Medical countermeasures and biodefense chapter completed","change":"Medical Countermeasures and Biodefense gets full content covering the Strategic National Stockpile, BARDA's role bridging development gaps, vaccine development timelines, Operation Warp Speed, and Emergency Use Authorization and Animal Rule regulatory pathways.","why":"Gives readers the handbook's core chapter on U.S. biodefense stockpiling and countermeasure development.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[]},{"category":"content","title":"Synthetic biology chapter added; countermeasures chapter expanded","change":"Synthetic Biology and the Democratization of Biotechnology covers enabling technologies, falling DNA synthesis costs, BioBricks, the iGEM competition, and how synthesis orders get flagged. Medical Countermeasures and Biodefense treats PPE allocation, antibody effectiveness against variants, and vaccine equity.","why":"Extends the handbook into synthetic biology's declining barriers to entry and adds real-world friction points to the countermeasures chapter.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[]}]},
{"date":"2025-12-04","headline":"Five chapters completed, from DNA synthesis screening to AI risk amplification","entries":[{"category":"content","title":"DNA synthesis screening chapter completed","change":"DNA Synthesis Screening gets full content on the two-layer sequence and customer screening approach, the IGSC Harmonized Screening Protocol, the 2010 and 2023 HHS screening guidance, and gaps such as short fragments and benchtop synthesizers that escape current screening.","why":"Gives readers a detailed account of the handbook's central choke-point intervention for biosecurity.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Federal Register, HHS biosecurity screening framework, 2010","url":"https://www.federalregister.gov/documents/2010/11/29/2010-29919/biosecurity-of-laboratory-facilities","type":"policy"},{"label":"HHS, updated nucleic acid synthesis screening framework, 2023","url":"https://www.hhs.gov/about/news/2023/10/20/hhs-releases-updated-framework-nucleic-acid-synthesis-screening.html","type":"policy"},{"label":"NTI, Common Mechanism project","url":"https://www.nti.org/about/programs-projects/project/biosecurity-innovation-and-risk-reduction/","type":"report"},{"label":"IBBIS, Know Your Customer","url":"https://ibbis.bio","type":"report"},{"label":"NIH, framework for nucleic acid synthesis screening","url":"https://www.nih.gov/about-nih/what-we-do/science-health-public-trust/perspectives/science-policy/framework-nucleic-acid-synthesis-screening","type":"policy"},{"label":"bioRxiv, BLAST-based sequence screening, 2024, preprint","url":"https://www.biorxiv.org/content/10.1101/2024.03.07.583915v1","type":"preprint"}]},{"category":"content","title":"Gain-of-function research chapter completed","change":"Gain-of-Function Research covers the 2011-2012 H5N1 ferret transmissibility studies, the 2014-2017 U.S. funding pause, the 2017 HHS P3CO framework, and the May 2024 dual-use and enhanced pandemic potential oversight policy, noting its implementation was paused in May 2025 pending review.","why":"Traces the science and shifting U.S. oversight of gain-of-function research from its defining controversy to current policy.","chapters":[["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"]],"sources":[{"label":"Fouchier et al., Science, 2012","url":"https://www.science.org/doi/10.1126/science.1213362","type":"paper"},{"label":"Kawaoka et al., Nature, 2012","url":"https://www.nature.com/articles/nature10831","type":"paper"},{"label":"HHS, gain-of-function funding pause announcement, 2014","url":"https://www.hhs.gov/about/news/2014/10/17/us-government-gain-function-deliberative-process-and-research-funding-pause-potential-pandemic-pathogens.html","type":"policy"},{"label":"HHS, P3CO Framework, 2017","url":"https://www.hhs.gov/about/news/2017/12/19/hhs-p3co-framework.html","type":"policy"},{"label":"HHS, DURC and Pathogens with Enhanced Pandemic Potential policy, 2024","url":"https://www.hhs.gov/about/news/2024/05/08/hhs-announces-new-united-states-government-policy-oversight-dual-use-research-concern-pathogens-enhanced-pandemic-potential.html","type":"policy"},{"label":"CIDRAP, H5N1 avian influenza overview","url":"https://www.cidrap.umn.edu/avian-influenza-bird-flu/h5n1-influenza-virus","type":"news"}]},{"category":"content","title":"Gene drives chapter completed","change":"Gene Drives and Environmental Biosecurity explains CRISPR-based gene drives and super-Mendelian inheritance, covers applications such as Target Malaria's mosquito drives, and assesses ecological risks and governance gaps under the Convention on Biological Diversity and Cartagena Protocol.","why":"Extends the handbook's emerging-technology coverage to a dual-use tool for population-level genetic engineering.","chapters":[["Gene Drives and Environmental Biosecurity","/biotechnology/gene-drives.html"]],"sources":[{"label":"Burt, Proceedings of the Royal Society B, 2003","url":"https://royalsocietypublishing.org/doi/10.1098/rspb.2002.2319","type":"paper"},{"label":"Target Malaria","url":"https://targetmalaria.org/","type":"other"},{"label":"Convention on Biological Diversity","url":"https://www.cbd.int/","type":"policy"},{"label":"Cartagena Protocol on Biosafety","url":"https://bch.cbd.int/protocol","type":"policy"},{"label":"WHO, malaria fact sheet","url":"https://www.who.int/news-room/fact-sheets/detail/malaria","type":"policy"},{"label":"BMC Biology, gene drive modeling study","url":"https://bmcbiol.biomedcentral.com/articles/10.1186/s12915-020-00875-0","type":"paper"}]},{"category":"content","title":"AI as a biosecurity risk amplifier chapter completed","change":"The AI as a Biosecurity Risk Amplifier chapter distinguishes language-model information hazards from biological design tool design hazards, reviews the 2022 MegaSyn toxic-molecule generation experiment, and weighs OpenAI's and Anthropic's uplift evaluations against the tacit-knowledge barrier to bioweapons acquisition.","why":"Frames how the handbook separates AI text-generation risks from AI biological-design risks for readers assessing either.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"OpenAI, early warning system for biological threats, 2024","url":"https://cdn.openai.com/papers/building_an_early_warning_system_for_llm_aided_biological_threat_creation.pdf","type":"announcement"},{"label":"Anthropic, Claude model card, 2024","url":"https://www.anthropic.com/research/claude-model-card","type":"announcement"},{"label":"OSTP, framework for nucleic acid synthesis screening, 2024","url":"https://www.whitehouse.gov/ostp/news-updates/2024/04/29/framework-for-nucleic-acid-synthesis-screening/","type":"policy"},{"label":"Executive Order 14110, safe and trustworthy AI, 2023","url":"https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/","type":"policy"},{"label":"Sandbrink, dual-use AI biosecurity analysis, 2023, preprint","url":"https://arxiv.org/abs/2306.13952","type":"preprint"},{"label":"Amodei, congressional testimony, 2023","url":"https://www.youtube.com/watch?v=AaSzb7aRWQE","type":"other"}]},{"category":"content","title":"LLMs and information hazards chapter completed","change":"The LLMs and Information Hazards chapter assesses what language models can and cannot do for biological attack planning, covers their tendency to state incorrect information confidently, and reviews the RAND Corporation's 2024 red-team study alongside jailbreaking research and reinforcement learning from human feedback as a safeguard.","why":"Separates demonstrated language-model limitations from information already accessible elsewhere, a recurring question in AI biosecurity debates.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Adversarial prompts and jailbreaking study, 2023, preprint","url":"https://arxiv.org/abs/2307.15043","type":"preprint"},{"label":"Asilomar Conference on recombinant DNA safety, 1975","url":"https://www.nature.com/articles/nbt1207-1265","type":"paper"},{"label":"Gryphon Scientific, biological attack barriers study for USAID","url":"https://www.gryphonscientific.com/","type":"other"},{"label":"Reinforcement learning from human feedback, preprint","url":"https://arxiv.org/abs/1706.03741","type":"preprint"},{"label":"Anthropic, Constitutional AI","url":"https://www.anthropic.com/research/constitutional-ai","type":"announcement"},{"label":"Frontier Model Forum","url":"https://www.frontiermodelforum.org/","type":"report"}]}]},
{"date":"2025-12-05","headline":"AI-Enabled Pathogen Design chapter; AI for Biosecurity Defense chapter; Cloud Labs and Automated Biology chapter; LLMs and Information Hazards: defining hazard types; Policy Frameworks and Biosecurity Careers chapters; Attribution and Biological Forensics chapter; Case Studies in Biosecurity Failures and Successes chapter; The Future of Biosecurity chapter; Glossary, chapter summaries, and a Further Reading list; Pathogen genomics for precision public health","entries":[{"category":"content","title":"AI-Enabled Pathogen Design chapter","change":"Adds a full chapter on generative protein and pathogen design tools, covering AlphaFold-era structure prediction, generative design methods such as RFdiffusion and ESM3, the NASEM 2024 three-tier risk framework, and the 2022 Urbina experiment.","why":"Grounds the chapter's risk assessment in primary sources: the Nobel-recognized protein-design advances and the NASEM 2024 three-tier framework.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[{"label":"The Nobel Prize in Chemistry, press release, 2024","url":"https://www.nobelprize.org/prizes/chemistry/2024/press-release/","type":"other"},{"label":"National Academies (NASEM), 2024","url":"https://doi.org/10.17226/28868","type":"paper"},{"label":"Urbina et al., 2022","url":"https://doi.org/10.1038/s42256-022-00465-9","type":"paper"}]},{"category":"content","title":"AI for Biosecurity Defense chapter","change":"Adds a full chapter on defensive AI applications: outbreak anomaly detection platforms such as BlueDot, metagenomic pathogen identification through efforts like the Nucleic Acid Observatory, and AI-accelerated countermeasure development.","why":"Gives readers primary sources for the chapter's defensive use cases in surveillance, pathogen detection, and countermeasure design.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"BlueDot","url":"https://bluedot.global/","type":"other"},{"label":"Nucleic Acid Observatory","url":"https://www.naobservatory.org/","type":"other"},{"label":"Nature News, on Moderna's mRNA platform","url":"https://www.nature.com/articles/d41586-021-02483-w","type":"news"}]},{"category":"content","title":"Cloud Labs and Automated Biology chapter","change":"Adds a full chapter on remote, automated laboratories, describing commercial platforms such as Emerald Cloud Lab and Strateos and a RAND analysis of the biosecurity assumptions cloud labs disrupt.","why":"Grounds claims about remote experiment access in named commercial platforms and RAND's assessment of the traditional biosecurity model.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"Emerald Cloud Lab","url":"https://www.emeraldcloudlab.com/","type":"other"},{"label":"Strateos","url":"https://strateos.com/","type":"other"},{"label":"RAND, perspective on cloud lab biosecurity","url":"https://www.rand.org/pubs/perspectives/PEA3851-1.html","type":"report"}]},{"category":"content","title":"LLMs and Information Hazards: defining hazard types","change":"Expands the chapter with Bostrom's typology of information hazards, including signal hazards illustrated by the 2018 horsepox synthesis study, a working definition of biological 'uplift,' and an updated Gryphon Scientific assessment covering 2023 to 2024.","why":"Gives readers a working vocabulary for information hazards and grounds the uplift debate in Bostrom's original framework and Gryphon Scientific's assessments.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Bostrom, Information Hazards","url":"https://nickbostrom.com/information-hazards.pdf","type":"other"},{"label":"2018 horsepox synthesis study, PLOS ONE","url":"https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0188453","type":"paper"},{"label":"Esvelt and colleagues, video commentary","url":"https://www.youtube.com/watch?v=Y5Xv-rmmsmY","type":"post"}]},{"category":"content","title":"Policy Frameworks and Biosecurity Careers chapters","change":"Adds two new chapters: policy frameworks for AI-biological risk, covering the EU AI Act, Anthropic's Responsible Scaling Policy, and the Bletchley Declaration's AI Safety Institute network; and biosecurity career pathways, mapping entry routes and fellowships such as the Emerging Leaders in Biosecurity program.","why":"Gives readers primary sources for the governance frameworks shaping AI-biological risk policy, and names concrete entry points into the biosecurity workforce.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Building a Biosecurity Career: Federal Agencies, Fellowships, and AI-Bio Roles","/governance/biosecurity-careers.html"]],"sources":[{"label":"EU AI Act, high-level summary","url":"https://artificialintelligenceact.eu/high-level-summary/","type":"other"},{"label":"Anthropic, Responsible Scaling Policy","url":"https://www.anthropic.com/news/anthropics-responsible-scaling-policy","type":"announcement"},{"label":"UK Government, Bletchley Declaration, 2023","url":"https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration","type":"policy"},{"label":"Johns Hopkins Center for Health Security, Emerging Leaders in Biosecurity","url":"https://www.centerforhealthsecurity.org/education-training/elbi/","type":"report"}]},{"category":"content","title":"Attribution and Biological Forensics chapter","change":"Adds a full chapter on microbial forensics and attribution science, covering the FBI's Amerithrax whole-genome-sequencing investigation, the DHS National Bioforensic Analysis Center, and a RAND assessment of attribution's role in deterrence.","why":"Grounds the chapter's forensic methods and deterrence claims in the FBI's own investigation record, DHS's forensic center, and independent RAND analysis.","chapters":[["Attribution and Biological Forensics","/governance/attribution-forensics.html"]],"sources":[{"label":"FBI, Amerithrax investigation","url":"https://www.fbi.gov/investigate/wmd/amerithrax-investigation","type":"policy"},{"label":"RAND, report on biological weapons attribution","url":"https://www.rand.org/pubs/research_reports/RRA2360-1.html","type":"report"},{"label":"DHS, National Bioforensic Analysis Center","url":"https://www.dhs.gov/science-and-technology/nbfac","type":"policy"}]},{"category":"content","title":"Case Studies in Biosecurity Failures and Successes chapter","change":"Adds a full chapter of historical case studies, including Soviet Biopreparat, the Amerithrax attribution, the Sverdlovsk 1979 incident ('the Biological Chernobyl'), the 1977 H1N1 lab reemergence, and Aum Shinrikyo's failed bioterrorism attempts, with cross-cutting lessons for practitioners.","why":"Gives readers documented sources for the field's core historical lessons, including Soviet program analysis and the FBI's own Amerithrax investigative record.","chapters":[["Case Studies in Biosecurity Failures and Successes","/governance/case-studies.html"]],"sources":[{"label":"NTI, Soviet Union's offensive biological weapons program","url":"https://www.nti.org/analysis/articles/soviet-unions-offensive-biological-weapons-program/","type":"report"},{"label":"US Department of Justice, Amerithrax investigative summary","url":"https://www.justice.gov/archive/amerithrax/docs/amx-investigative-summary.pdf","type":"policy"},{"label":"New York Times, on Ken Alibek's account, 1998","url":"https://www.nytimes.com/1998/02/25/world/soviet-defector-warns-of-biological-weapons.html","type":"news"}]},{"category":"content","title":"The Future of Biosecurity chapter","change":"Adds a closing chapter on emerging trends and future scenarios for biological risk, citing CEPI's 100 Days Mission vaccine-development target and a World Bank benchmark for pandemic preparedness financing.","why":"Gives readers the primary targets behind pandemic-preparedness benchmarks referenced in the chapter's outlook for the field.","chapters":[["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[{"label":"CEPI, 100 Days Mission","url":"https://cepi.net/100-days/","type":"other"},{"label":"World Bank, Pandemic Fund","url":"https://www.worldbank.org/en/topic/pandemics/publication/pandemic-fund","type":"other"}]},{"category":"content","title":"Glossary, chapter summaries, and a Further Reading list","change":"The appendices hold a glossary spanning the alphabet, a chapter-by-chapter TL;DR compilation, and a Further Reading list of books, reports, and policy documents. Case studies and career pathways each have a full chapter of their own.","why":"Gives readers a single glossary, a rapid per-chapter summary, and a curated reading list.","chapters":[["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"OSTP, AI biodefense framework, 2024","url":"https://www.whitehouse.gov/ostp/","type":"policy"},{"label":"HHS, P3CO policy","url":"https://www.phe.gov/s3/dualuse/Pages/p3co.aspx","type":"policy"},{"label":"WHO, Laboratory Biosafety Manual, 4th edition","url":"https://www.who.int/publications/i/item/9789240011311","type":"policy"}]},{"category":"evidence","title":"Pathogen genomics for precision public health","change":"Outbreak Detection and Surveillance adds a section on pathogen genomics, describing real-time sequencing's role in precision public health surveillance.","why":"Cites the New England Journal of Medicine review as the primary source for this emerging surveillance approach.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"Armstrong et al., New England Journal of Medicine","url":"https://www.nejm.org/doi/full/10.1056/NEJMsr1813907","type":"paper"}]}]},
{"date":"2025-12-09","headline":"COVID-19 origins discussion revised","entries":[{"category":"revision","title":"COVID-19 origins discussion revised","change":"The Biological Threat Landscape now states that scientific consensus supports natural zoonotic spillover, consistent with SARS-CoV-1 and MERS-CoV precedent, notes that laboratory-origin hypotheses lack direct supporting evidence, and observes that the debate has become politically charged.","why":"Situates the chapter's origins discussion in the scientific consensus while acknowledging why the debate remains contested.","chapters":[["The Biological Threat Landscape","/foundations/threat-landscape.html"]],"sources":[]}]},
{"date":"2025-12-10","headline":"Benchtop synthesis and tacit-knowledge barriers reassessed; Time-bound LLM uplift findings and multimodal capabilities; DNA Synthesis Screening: benchtop bypass problem; BWC complementary frameworks and H5N1 dairy cattle case study","entries":[{"category":"revision","title":"Benchtop synthesis and tacit-knowledge barriers reassessed","change":"DNA Synthesis Screening now frames benchtop DNA synthesizers as the most serious long-term threat to the screening chokepoint strategy, and AI as a Biosecurity Risk Amplifier adds a caveat that emerging multimodal AI may erode the tacit-knowledge barrier that text-based models cannot bridge.","why":"Sharpens two of the handbook's core threat assessments as the underlying technology changes: chokepoint screening and the hands-on skills barrier.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[]},{"category":"revision","title":"Time-bound LLM uplift findings and multimodal capabilities","change":"AI as a Biosecurity Risk Amplifier and LLMs and Information Hazards add a 'snapshots in time' caveat to current 'mild uplift' findings, update their model examples to GPT-5.1, Claude Opus 4.5, and Gemini 3 Pro, and note that vision-enabled multimodal AI may erode the tacit-knowledge barrier faster than assumed.","why":"Keeps the uplift and tacit-knowledge assessments current: today's reassuring findings are time-bound, and multimodal AI is named as a specific reason the skills barrier may narrow.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[]},{"category":"content","title":"DNA Synthesis Screening: benchtop bypass problem","change":"DNA Synthesis Screening adds sections on benchtop synthesizer technologies, the screening bypass problem these devices create, policy and governance gaps, and the NTI Common Mechanism as a potential solution, naming platforms from DNA Script, Ansa Biotechnologies, Nuclera, and Twist Bioscience.","why":"Names the specific commercial platforms and proposed governance mechanism behind the chapter's benchtop-synthesis gap analysis.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"]],"sources":[{"label":"DNA Script","url":"https://www.dnascript.com/","type":"other"},{"label":"Ansa Biotechnologies","url":"https://www.ansabio.com/","type":"other"},{"label":"Nuclera","url":"https://nuclera.com/","type":"other"},{"label":"Twist Bioscience","url":"https://www.twistbioscience.com/","type":"announcement"}]},{"category":"content","title":"BWC complementary frameworks and H5N1 dairy cattle case study","change":"The Biological Weapons Convention chapter adds a section on complementary international frameworks (UN Security Council Resolution 1540, the Australia Group, China's 2021 Biosecurity Law), What Is Biosecurity? adds a One Health section, and Case Studies adds an ongoing account of H5N1 in U.S. dairy cattle since 2024.","why":"Extends the treaty-governance chapter to overlapping instruments, and adds a live example of One Health surveillance gaps: the still-developing H5N1 dairy cattle situation.","chapters":[["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"],["What Is Biosecurity?","/foundations/what-is-biosecurity.html"],["Case Studies in Biosecurity Failures and Successes","/governance/case-studies.html"]],"sources":[{"label":"WHO, One Health","url":"https://www.who.int/news-room/questions-and-answers/item/one-health","type":"policy"},{"label":"UN, Security Council Resolution 1540","url":"https://www.un.org/disarmament/wmd/sc1540/","type":"policy"},{"label":"1540 Committee","url":"https://www.un.org/en/sc/1540/","type":"policy"}]}]},
{"date":"2025-12-11","headline":"Executive Summary published; What We Know framework added to six chapters","entries":[{"category":"milestone","title":"Executive Summary published","change":"Adds a whole-book Executive Summary: key findings on classical biosecurity gaps and fragmented international governance, priority recommendations for four audiences, a threat-prioritization table, and an implementation timeline. Same-day refinements add an opening paragraph and update AI-model references to GPT-5.1, Claude Opus 4.5, and Gemini 3 Pro.","why":"Gives readers a single, navigable synthesis of the handbook's findings and recommendations, without needing to read every chapter.","chapters":[["Executive Summary","/executive-summary.html"]],"sources":[]},{"category":"content","title":"'What We Know vs. What Remains Uncertain' framework added","change":"Six chapters, spanning gain-of-function, gene drives, and the AI-biosecurity chapters, add a 'What We Know vs. What Remains Uncertain' section that separates demonstrated findings from contested, theoretical, and unknown claims; the Preface adds a note on serving both biosecurity practitioners and AI safety researchers.","why":"Gives readers a consistent way to gauge how solid each chapter's claims are, and names the two audiences the handbook is written to serve.","chapters":[["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"],["Gene Drives and Environmental Biosecurity","/biotechnology/gene-drives.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"],["Preface","/preface.html"]],"sources":[]}]},
{"date":"2025-12-12","headline":"Multimodal AI's erosion of hands-on lab skill barriers; AI-designed sequences evade DNA synthesis screening; RAND analyses on AI-biological risk, synthesis screening, and DURC review; Fink Report origins of dual-use research oversight","entries":[{"category":"revision","title":"Multimodal AI's erosion of hands-on lab skill barriers","change":"Refines text in ten chapters: expands on vision-enabled AI models providing real-time laboratory coaching in the Executive Summary, AI as a Biosecurity Risk Amplifier, and LLMs and Information Hazards chapters, and updates framing in chapters on Biological Weapons Convention governance, medical countermeasures, synthetic biology, gain-of-function research, and case studies.","why":"Sharpens how the handbook frames eroding hands-on skill barriers as multimodal AI models advance, and tightens supporting detail in adjacent chapters.","chapters":[["Executive Summary","/executive-summary.html"],["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"],["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Case Studies in Biosecurity Failures and Successes","/governance/case-studies.html"]],"sources":[]},{"category":"evidence","title":"AI-designed sequences evade DNA synthesis screening","change":"DNA Synthesis Screening adds an October 2025 demonstration that AI-designed sequences evaded homology-based screening (patches now catch about 97%), and traces screening policy from 2010 HHS guidance through the April 2024 OSTP Framework to the May 2025 Executive Order 14292, whose revision deadline passed without a replacement.","why":"Gives readers the primary evidence that AI-designed sequences can elude homology-based screening, and explains why the policy meant to close such gaps has stalled.","chapters":[["Executive Summary","/executive-summary.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Wittmann et al., Science, 2025","url":"https://www.science.org/doi/10.1126/science.adu8578","type":"paper"},{"label":"OSTP, Framework for Nucleic Acid Synthesis Screening, 2024","url":"https://aspr.hhs.gov/S3/Pages/OSTP-Framework-for-Nucleic-Acid-Synthesis-Screening.aspx","type":"policy"},{"label":"Executive Order 14292, 2025","url":"https://www.whitehouse.gov/presidential-actions/2025/05/improving-the-safety-and-security-of-biological-research/","type":"policy"},{"label":"IBBIS, Common Mechanism, 2025","url":"https://ibbis.bio/our-work/common-mechanism/","type":"report"},{"label":"International Organization for Standardization, ISO 20688-2:2024","url":"https://www.iso.org/standard/82932.html","type":"other"}]},{"category":"evidence","title":"RAND analyses on AI-biological risk, synthesis screening, and DURC review","change":"Adds five RAND reports to Further Reading and Resources covering AI-enabled biological tools, synthesis screening, and pathogen design, and integrates RAND analysis into chapters on dual-use research review, wastewater surveillance funding, AI risk amplification, AI-enabled pathogen design, and policy frameworks for AI-bio convergence.","why":"Adds primary RAND research the handbook had not yet cited on AI-biological risk, synthesis screening policy, and dual-use research oversight.","chapters":[["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"RAND, Protecting Biological Materials from Misuse, 2025","url":"https://www.rand.org/pubs/research_reports/RRA4067-1.html","type":"report"},{"label":"RAND, Mitigating AI-CBW Risks, 2025","url":"https://www.rand.org/pubs/research_reports/RRA2990-1.html","type":"report"},{"label":"RAND, Theoretical Limits of AI-Enabled Pathogen Design, 2025","url":"https://www.rand.org/pubs/research_reports/RRA4087-1.html","type":"report"},{"label":"RAND, Securing Commercial Nucleic Acid Synthesis, 2024","url":"https://www.rand.org/pubs/research_reports/RRA3329-1.html","type":"report"},{"label":"RAND, Global Risk Index for AI-enabled Biological Tools, 2025","url":"https://www.rand.org/pubs/external_publications/EP71093.html","type":"report"},{"label":"RAND Meselson Center, dual-use research decisionmaking tool, 2025","url":"https://www.rand.org/pubs/perspectives/PEA4125-1.html","type":"report"}]},{"category":"content","title":"Fink Report origins of dual-use research oversight","change":"Dual-Use Research of Concern adds a Historical Origins section on the 2004 Fink Report and its seven categories of experiments of concern, with a matching Glossary entry; Gain-of-Function Research clarifies that Executive Order 14292 paused the 2024 gain-of-function policy before its May 6, 2025 effective date, leaving the interim HHS P3CO Framework in effect.","why":"Grounds the DURC framework's origin for readers and clarifies which gain-of-function policy currently applies while the 2025 revision remains pending.","chapters":[["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"]],"sources":[]}]},
{"date":"2025-12-15","headline":"Two new chapters: digital biosurveillance and surveillance equity","entries":[{"category":"content","title":"Two new chapters: digital biosurveillance and surveillance equity","change":"Adds two chapters: Digital Biosurveillance, covering wearable-based outbreak detection and AI-enabled genomic surveillance pipelines like Nextstrain and Pangolin, and Global Surveillance Equity, on WHO capacity assessments and the incentive gap illustrated by South Africa's Omicron reporting.","why":"Gives readers dedicated coverage of emerging outbreak-detection technologies and the incentives that discourage transparent reporting from under-resourced countries.","chapters":[["Digital Biosurveillance: From Wearables to Real-Time Genomics","/ai-biosecurity/digital-biosurveillance.html"],["Global Surveillance Equity: LMIC Capacity as Collective Security","/governance/surveillance-equity.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"],["Preface","/preface.html"]],"sources":[{"label":"Scripps Research, DETECT wearable-detection study","url":"https://doi.org/10.1038/s41591-020-1123-x","type":"paper"},{"label":"Stanford Medicine, smartwatch COVID-19 detection study","url":"https://doi.org/10.1038/s41551-020-00640-6","type":"paper"},{"label":"Nextstrain, real-time pathogen phylogenetics platform","url":"https://doi.org/10.1093/bioinformatics/bty407","type":"paper"},{"label":"WHO, Joint External Evaluation framework","url":"https://www.who.int/emergencies/operations/international-health-regulations-monitoring-evaluation-framework/joint-external-evaluations","type":"policy"},{"label":"Africa Pathogen Genomics Initiative, Lancet Microbe, 2020","url":"https://www.thelancet.com/journals/lanmic/article/PIIS2666-5247(20)30206-8/fulltext","type":"paper"},{"label":"NPR, Guterres on Omicron travel bans, 2021","url":"https://www.npr.org/sections/coronavirus-live-updates/2021/11/28/1059619823/omicron-travel-bans-covid","type":"news"}]},{"category":"content","title":"Glossary entries for biosurveillance and benefit-sharing terms","change":"Adds five Glossary of Biosecurity Terms entries: Africa Pathogen Genomics Initiative, Benefit-sharing, Digital Biomarkers, Joint External Evaluation (JEE), and Nextstrain.","why":"Defines the terminology introduced by the new digital biosurveillance and surveillance equity chapters.","chapters":[["Glossary of Biosecurity Terms","/appendices/glossary.html"]],"sources":[]}]},
{"date":"2025-12-16","headline":"New section: national leadership as pandemic security","entries":[{"category":"content","title":"New section: national leadership as pandemic security","change":"The Future of Biosecurity adds a National Leadership as Pandemic Security section arguing preparedness succeeds or fails nationally, citing the Pandemic Fund's $1.385 billion in awards, the IMF's Resilience and Sustainability Trust, and WHO's HEPR framework.","why":"Makes the case that global financing mechanisms only work if countries sustain preparedness investment between crises.","chapters":[["The Future of Biosecurity","/governance/future-biosecurity.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"The Pandemic Fund, 2025","url":"https://www.thepandemicfund.org/","type":"other"},{"label":"IMF, Resilience and Sustainability Trust","url":"https://www.imf.org/en/Topics/Resilience-and-Sustainability-Trust","type":"other"},{"label":"WHO, HEPR framework, 2023","url":"https://www.who.int/publications/m/item/strengthening-the-global-architecture-for-health-emergency-prevention--preparedness--response-and-resilience","type":"policy"},{"label":"Van Kerkhove and Ihekweazu, 2025","url":"https://doi.org/10.1126/science.aee5227","type":"paper"}]}]},
{"date":"2025-12-17","headline":"Chatham House report on laboratory accidents and biocontainment breaches","entries":[{"category":"evidence","title":"Chatham House report on laboratory accidents and biocontainment breaches","change":"The Biological Threat Landscape cites a Chatham House report on laboratory accidents and biocontainment breaches to support examples including the 2004 SARS Beijing outbreak and 2014 CDC smallpox vial discovery.","why":"Grounds the handbook's discussion of underreported laboratory-acquired infections in a documented catalog of biocontainment incidents.","chapters":[["The Biological Threat Landscape","/foundations/threat-landscape.html"],["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"]],"sources":[{"label":"Chatham House, laboratory accidents and biocontainment breaches, 2023","url":"https://www.chathamhouse.org/2023/12/laboratory-accidents-and-biocontainment-breaches","type":"report"}]},{"category":"policy","title":"H.R. 3677: codifying DNA synthesis screening requirements","change":"AI as a Biosecurity Risk Amplifier updates its account of DNA synthesis screening: the 2024 OSTP Framework was rescinded by the May 2025 Executive Order 14292, which directs OSTP to develop a revised framework with enforcement mechanisms, and notes H.R. 3677 was introduced to codify these requirements into law.","why":"Tracks a pending congressional bill that would give the screening mandate a statutory basis after the 2024 framework's rescission.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"H.R. 3677, 119th Congress, 2025","url":"https://www.congress.gov/bill/119th-congress/house-bill/3677/text","type":"policy"}]}]},
{"date":"2025-12-19","headline":"GovAI framework for quantifying AI biosecurity risk; OpenAI reports GPT-5 wet-lab protocol optimization demonstration; Simplified phrasing on AI-biosecurity convergence","entries":[{"category":"evidence","title":"GovAI framework for quantifying AI biosecurity risk","change":"AI as a Biosecurity Risk Amplifier, LLMs and Information Hazards, and AI-Enabled Pathogen Design each add a 2025 GovAI framework (Luca Righetti) that converts capability-uplift findings into quantified risk estimates, illustrating how a 10 percentage point increase in capable actors could translate into population-level harm.","why":"Moves the discussion from whether AI provides uplift to how much that uplift could matter at a population scale.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[{"label":"Righetti, GovAI, converting capability evaluations to risk assessments, 2025","url":"https://www.governance.ai/research-paper/dual-use-ai-capabilities-and-the-risk-of-bioterrorism-converting-capability-evaluations-to-risk-assessments","type":"report"}]},{"category":"developments","title":"OpenAI reports GPT-5 wet-lab protocol optimization demonstration","change":"AI as a Biosecurity Risk Amplifier, LLMs and Information Hazards, and Cloud Labs and Automated Biology each add OpenAI's December 2025 demonstration of GPT-5 iterating on wet-lab experiments with biosecurity firm Red Queen Bio, reporting a 79-fold efficiency improvement in molecular cloning through AI-driven protocol optimization.","why":"Tracks developer-reported evidence that AI can now iterate on physical lab protocols, not just provide information, narrowing the hands-on skills gap.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"OpenAI, accelerating biological research in the wet lab, 2025","url":"https://openai.com/index/accelerating-biological-research-in-the-wet-lab/","type":"announcement"}]},{"category":"revision","title":"Simplified phrasing on AI-biosecurity convergence","change":"Simplifies phrasing describing where AI and biosecurity intersect in the Quick Reference compilation, the Glossary, AI as a Biosecurity Risk Amplifier, Policy Frameworks for AI-Bio Convergence, Building a Biosecurity Career, and The Future of Biosecurity.","why":"Tightens repeated phrasing across chapters without changing the underlying claims.","chapters":[["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Building a Biosecurity Career: Federal Agencies, Fellowships, and AI-Bio Roles","/governance/biosecurity-careers.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[]}]},
{"date":"2025-12-21","headline":"Quick Answer and FAQ sections added to foundational chapters; Quick Answer and FAQ sections added across emerging-risk, AI, and governance chapters","entries":[{"category":"content","title":"Quick Answer and FAQ sections added to foundational chapters","change":"Adds Quick Answer summaries to What Is Biosecurity?, Pathogens of Concern, Laboratory Biosafety and Biosecurity, Dual-Use Research of Concern, Outbreak Detection and Surveillance, and Medical Countermeasures and Biodefense, and Frequently Asked Questions sections to the TL;DR compilation and Further Reading and Resources appendices.","why":"Gives readers a fast, scannable summary before the full chapter discussion.","chapters":[["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"],["What Is Biosecurity?","/foundations/what-is-biosecurity.html"],["Pathogens of Concern","/foundations/pathogens-of-concern.html"],["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[]},{"category":"content","title":"Quick Answer and FAQ sections added across emerging-risk, AI, and governance chapters","change":"Adds Quick Answer summaries and Frequently Asked Questions sections to the emerging-risks chapters (synthetic biology, DNA synthesis screening, gain-of-function research, gene drives), the AI-biosecurity chapters, and the governance chapters, including specific questions such as what distinguishes AlphaFold from RFdiffusion and how BlueDot detected COVID-19 before WHO.","why":"Gives readers fast entry points into the handbook's most technical chapters through scannable summaries and common questions.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"],["Gene Drives and Environmental Biosecurity","/biotechnology/gene-drives.html"],["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"],["Attribution and Biological Forensics","/governance/attribution-forensics.html"],["Global Surveillance Equity: LMIC Capacity as Collective Security","/governance/surveillance-equity.html"]],"sources":[]}]},
{"date":"2025-12-24","headline":"Global Biofoundries Alliance linked in platform table","entries":[{"category":"revision","title":"Global Biofoundries Alliance linked in platform table","change":"Adds a link to the Global Biofoundries Alliance alongside the university biofoundries row in the Cloud Labs chapter's platform comparison table.","why":"Gives readers a primary reference for university biofoundry access.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"Global Biofoundries Alliance","url":"https://www.biofoundries.org/","type":"other"}]}]},
{"date":"2025-12-25","headline":"Mirror life biosecurity warnings added to synthetic biology chapter","entries":[{"category":"evidence","title":"Mirror life biosecurity warnings added to synthetic biology chapter","change":"Adds a new Synthetic Biology section on mirror life (mirror-image DNA and proteins), covering the December 2024 Science warning from 38 researchers and its immune evasion risks, plus glossary definitions for chirality and mirror life. The Biological Threat Landscape chapter adds two historical accidental-release case studies: the 1977 H1N1 'Russian Flu' and the 1978 Birmingham smallpox outbreak.","why":"Gives readers the primary scientific warning behind calls to halt mirror organism research, and documents past laboratory accidents relevant to today's biosafety debates.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["The Biological Threat Landscape","/foundations/threat-landscape.html"]],"sources":[{"label":"Science, December 2024","url":"https://www.science.org/content/article/leading-scientists-urge-ban-developing-mirror-image-bacteria","type":"news"},{"label":"Stanford News, December 2024","url":"https://news.stanford.edu/stories/2024/12/potential-risks-of-mirror-life","type":"news"},{"label":"Johns Hopkins Center for Health Security, 2024","url":"https://centerforhealthsecurity.org/2024/chs-director-joins-other-experts-in-science-commentary-to-highlight-potential-risks-of-mirror-bacteria","type":"report"},{"label":"Bulletin of the Atomic Scientists, 2024","url":"https://thebulletin.org/2024/12/scientists-urge-halt-to-research-on-creating-synthetic-mirror-bacteria-that-could-evade-human-immunity-disrupt-ecosystems/","type":"news"},{"label":"Think Global Health, 2024","url":"https://www.thinkglobalhealth.org/article/mirror-life-addressing-potential-biothreat","type":"other"}]}]},
{"date":"2025-12-26","headline":"Overview page added for the AI and Biosecurity part","entries":[{"category":"content","title":"Overview page added for the AI and Biosecurity part","change":"Adds a new landing page for the AI and Biosecurity part, summarizing its six chapters (risk amplification, information hazards, pathogen design, defense, digital biosurveillance, and red-teaming), key questions the section addresses, and suggested reading paths.","why":"Gives readers a single orientation point for the AI-biosecurity part before they choose a chapter.","chapters":[["AI and Machine Learning Fundamentals","/ai-biosecurity/index.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Welcome","/"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"],["Building a Biosecurity Career: Federal Agencies, Fellowships, and AI-Bio Roles","/governance/biosecurity-careers.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[]}]},
{"date":"2025-12-28","headline":"AI and machine learning fundamentals chapter added; WHO AI ethics guidance and AI-bio liability gap added to policy chapter; HHS AI Use Case Inventory added to Further Reading","entries":[{"category":"content","title":"AI and machine learning fundamentals chapter added","change":"Replaces the AI and Biosecurity part's overview page with a full chapter on AI and machine learning fundamentals for biosecurity readers, covering how neural networks and large language models work, base versus deployed models, safety training (RLHF and Constitutional AI), jailbreaking, confidently incorrect model output, and biological design tools versus general LLMs.","why":"Gives biosecurity readers without an AI background the conceptual grounding needed before the chapters on risk amplification and pathogen design.","chapters":[["AI and Machine Learning Fundamentals","/ai-biosecurity/index.html"]],"sources":[{"label":"3Blue1Brown, neural network series","url":"https://www.3blue1brown.com/topics/neural-networks","type":"other"},{"label":"Karpathy, AI tutorials","url":"https://karpathy.ai/","type":"other"},{"label":"Goodfellow et al., Deep Learning","url":"https://www.deeplearningbook.org/","type":"other"},{"label":"Vaswani et al., 2017, preprint","url":"https://arxiv.org/abs/1706.03762","type":"preprint"},{"label":"OpenAI, InstructGPT, 2022, preprint","url":"https://arxiv.org/abs/2203.02155","type":"preprint"}]},{"category":"policy","title":"WHO AI ethics guidance and AI-bio liability gap added to policy chapter","change":"Adds a new section to the Policy Frameworks chapter on the WHO's six ethical principles for AI in health, its value-chain framework for developers, providers, and deployers, and the absence of clear liability rules for AI-enabled biological harm. Adds an LMM (large multi-modal model) row to the AI terminology table, citing the same WHO guidance.","why":"Gives readers the primary WHO framework for AI governance responsibilities that the chapter's liability discussion builds on.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["AI and Machine Learning Fundamentals","/ai-biosecurity/index.html"]],"sources":[{"label":"WHO, 2024","url":"https://www.who.int/publications/i/item/9789240084759","type":"policy"}]},{"category":"policy","title":"HHS AI Use Case Inventory added to Further Reading","change":"Adds the HHS AI Use Case Inventory, a searchable catalog of AI applications across CDC, FDA, and NIH with 271 use cases as of 2024, to the Further Reading appendix's scientific databases section.","why":"Gives readers a primary source for tracking federal AI deployment in health agencies.","chapters":[["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"HHS, AI Use Case Inventory, 2024","url":"https://www.healthit.gov/hhs-ai-usecases","type":"policy"}]}]},
{"date":"2026-01-04","headline":"Target Malaria suspends Burkina Faso gene drive field activities; 2024 IHR amendments enter into force; DURC oversight policy update: EO 14292 pauses implementation; Nasal vaccine and far-UVC countermeasure sections added; Iraq biological weapons program case study added; Dissenting view on mirror bacteria immune evasion added; 2025 executive actions and state AI law status documented; Peaceful proxy evaluations and Fink Report AI-capability mapping added","entries":[{"category":"developments","title":"Target Malaria suspends Burkina Faso gene drive field activities","change":"Gene Drives and Environmental Biosecurity adds an update noting that Burkina Faso's Ministry of Higher Education, Research and Innovation suspended all Target Malaria activities on August 22, 2025, after a small-scale release of non-gene-drive genetically modified male mosquitoes.","why":"Documents a real-world suspension of a field gene drive program relevant to the chapter's deployment-timeline discussion.","chapters":[["Gene Drives and Environmental Biosecurity","/biotechnology/gene-drives.html"]],"sources":[{"label":"Target Malaria, activities suspended in Burkina Faso, 2025","url":"https://targetmalaria.org/latest/news/target-malaria-activities-suspended-in-burkina-faso/","type":"announcement"}]},{"category":"policy","title":"2024 IHR amendments enter into force","change":"Outbreak Detection and Surveillance adds a section on the 2024 International Health Regulations amendments, adopted by the 77th World Health Assembly through resolution WHA77.17 and in force since September 19, 2025, strengthening equity, capacity-building, and financing provisions.","why":"Documents a binding update to the international surveillance framework the chapter covers.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"World Health Assembly, resolution WHA77.17, 2024","url":"https://apps.who.int/gb/ebwha/pdf_files/WHA77/A77_R17-en.pdf","type":"policy"}]},{"category":"policy","title":"DURC oversight policy update: EO 14292 pauses implementation","change":"Dual-Use Research of Concern (DURC) adds a Policy Framework Update noting that Executive Order 14292 (signed May 5, 2025) paused the May 2024 USG DURC oversight policy pending a 90-day review, and that the deadline passed without a replacement framework.","why":"Tracks the unsettled status of DURC oversight policy after the paused 2024 framework.","chapters":[["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"sources":[{"label":"HHS, USG Policy for Oversight of DURC and Pathogens with Enhanced Pandemic Potential, 2024","url":"https://www.phe.gov/s3/dualuse/Pages/oversight-durc-pathogens.aspx","type":"policy"}]},{"category":"content","title":"Nasal vaccine and far-UVC countermeasure sections added","change":"Medical Countermeasures and Biodefense adds sections on nasal vaccines and mucosal immunity, including Project NextGen's funding of Phase 1 to 2 trials, and on far-UVC environmental countermeasures, citing chamber studies showing 98 to 99% pathogen reduction while noting cluster-randomized trials are not yet complete.","why":"Expands biodefense coverage with two emerging countermeasure technologies not previously covered.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[{"label":"NIH, nasal COVID-19 vaccine trial, 2025","url":"https://www.nih.gov/news-events/news-releases/nih-sponsored-trial-nasal-covid-19-vaccine-opens","type":"announcement"},{"label":"Eadie et al., 2022","url":"https://doi.org/10.1038/s41598-022-08462-z","type":"paper"},{"label":"NIAID, Project NextGen","url":"https://www.niaid.nih.gov/diseases-conditions/next-generation-covid-19-vaccines","type":"announcement"}]},{"category":"content","title":"Iraq biological weapons program case study added","change":"Case Studies adds a case study on Iraq's biological weapons program (1974 to 1991), covering its concealment from UNSCOM inspectors for four years after the Gulf War, and compares it with the Soviet biological weapons program.","why":"Adds a historical case of a concealed weapons program to the chapter's comparative case studies.","chapters":[["Case Studies in Biosecurity Failures and Successes","/governance/case-studies.html"]],"sources":[{"label":"Duelfer Report, 2004","url":"https://www.govinfo.gov/app/details/GPO-DUELFERREPORT/","type":"report"},{"label":"Federation of American Scientists, citing UNSCOM","url":"https://nuke.fas.org/guide/iraq/bw/program.htm","type":"other"}]},{"category":"evidence","title":"Dissenting view on mirror bacteria immune evasion added","change":"Adds a glycobiologist's Science eLetter response questioning assumptions about mirror bacteria evading the immune system, noting existing cross-chiral immune recognition evidence, to the synthetic biology chapter's mirror life discussion.","why":"Adds scientific pushback on a widely cited biosecurity concern about synthetic mirror-image organisms.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"]],"sources":[{"label":"Derda et al., Science eLetter, 2025","url":"https://www.science.org/doi/10.1126/science.ads9158#elettersSection","type":"paper"}]},{"category":"policy","title":"2025 executive actions and state AI law status documented","change":"Adds a section on Executive Order 14292 (May 2025), which directed OSTP to add enforcement mechanisms to the voluntary synthesis screening framework, the AI Action Plan's split-order detection provisions, and a note that California SB 53 faces a federal preemption challenge under a December 2025 executive order.","why":"Tracks a shift from voluntary to enforceable synthesis screening rules and the federal-state tension over AI compute regulation.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Federal Register, Executive Order 14292, 2025","url":"https://www.federalregister.gov/documents/2025/05/08/2025-08266/improving-the-safety-and-security-of-biological-research","type":"policy"},{"label":"White House, America's AI Action Plan, 2025","url":"https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf","type":"policy"},{"label":"Federal Register, Executive Order 14179, 2025","url":"https://www.federalregister.gov/documents/2025/01/31/2025-02172/removing-barriers-to-american-leadership-in-artificial-intelligence","type":"policy"},{"label":"Federal Register, executive order on a national AI policy framework, 2025","url":"https://www.federalregister.gov/documents/2025/12/16/2025-23092/ensuring-a-national-policy-framework-for-artificial-intelligence","type":"policy"}]},{"category":"evidence","title":"Peaceful proxy evaluations and Fink Report AI-capability mapping added","change":"Red-Teaming AI Systems adds a section on peaceful proxy evaluations, wet-lab validation alternatives such as the Evo 2 genomic foundation model, following a U.S. government solicitation for such methods, and Dual-Use Research of Concern maps the Fink Report's seven categories of experiments of concern to emerging AI capabilities using the same underlying paper.","why":"Gives evaluators alternatives to live-agent testing and a shared framework for assessing AI-enabled biological capabilities against established dual-use categories.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"sources":[{"label":"U.S. government, peaceful proxy evaluations solicitation, 2025","url":"https://sam.gov/opp/03797b011bbb4a75ac925ac96a43f926/view","type":"policy"},{"label":"Pannu et al., 2025","url":"https://doi.org/10.1371/journal.pcbi.1012975","type":"paper"},{"label":"Evo 2 genomic foundation model, 2025, preprint","url":"https://www.biorxiv.org/content/10.1101/2025.02.18.638918v1","type":"preprint"}]}]},
{"date":"2026-01-06","headline":"Cloud labs, attribution, and biological threat detection","entries":[{"category":"revision","title":"Cloud labs, attribution, and biological threat detection","change":"Cloud-lab coverage notes that wet-lab automation still requires human troubleshooting. Synthetic biology coverage examines engineering signatures in attribution, alongside current biological threat detection capability.","why":"Describes technical capabilities and limitations in cloud labs, attribution, and biosecurity detection.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"],["Attribution and Biological Forensics","/governance/attribution-forensics.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[]}]},
{"date":"2026-01-08","headline":"Red-teaming chapter adds UK AISI, NCSC, and Sandbrink 2023 uplift evidence","entries":[{"category":"evidence","title":"Uplift studies and differential capability evidence expanded","change":"Expands the uplift studies discussion with UK AI Security Institute Chem-Bio Uplift Studies evidence, a Sandbrink 2023 forecast that AI could support novices acquiring biological weapons by late 2025, and the UK NCSC's assessment that AI uplift differs by actor type, applying that framework to biological threats.","why":"Documents government and academic evidence on whether and how AI narrows the gap between novice and expert biological capability.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"UK AI Security Institute, research","url":"https://www.aisi.gov.uk/research","type":"report"},{"label":"UK AI Security Institute, Chem-Bio Uplift Studies","url":"https://www.aisi.gov.uk/blog/advanced-ai-evaluations-may-update","type":"report"},{"label":"UK AI Security Institute, early lessons from evaluating frontier AI systems","url":"https://www.aisi.gov.uk/blog/early-lessons-from-evaluating-frontier-ai-systems","type":"report"},{"label":"National Cyber Security Centre, impact of AI on cyber threat, 2024","url":"https://www.ncsc.gov.uk/report/impact-of-ai-on-cyber-threat","type":"report"}]}]},
{"date":"2026-01-10","headline":"Anthropic's Constitutional Classifiers added as AI jailbreak defense","entries":[{"category":"developments","title":"Anthropic's Constitutional Classifiers added as jailbreak defense","change":"Adds sections on Anthropic's Constitutional Classifiers, which monitor AI exchanges in real time using classifiers trained on a natural-language 'constitution,' describing the January 2026 Constitutional Classifiers++ update, a developer-reported 0.05% false positive rate, and over 1,700 hours of red-teaming in which no universal jailbreak was found.","why":"Anthropic released a new generation of its jailbreak defense system with developer-reported production performance and red-teaming results.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"]],"sources":[{"label":"Constitutional Classifiers, 2026, preprint","url":"https://arxiv.org/abs/2601.04603","type":"preprint"},{"label":"Anthropic, next-generation Constitutional Classifiers, 2026","url":"https://www.anthropic.com/research/next-generation-constitutional-classifiers","type":"announcement"}]}]},
{"date":"2026-01-11","headline":"Global Risk Index for AI-enabled biological tools added","entries":[{"category":"evidence","title":"Global Risk Index for AI-enabled biological tools added","change":"Adds the Centre for Long-Term Resilience and RAND's Global Risk Index for AI-Enabled Biological Tools, a rubric assessing such tools on capability, misuse potential, accessibility, and technological maturity, to the red-teaming and policy frameworks chapters.","why":"Gives developers and funders a common framework to assess AI-enabled biological tools before development or deployment.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Centre for Long-Term Resilience and RAND, Global Risk Index for AI-Enabled Biological Tools, 2025","url":"https://www.longtermresilience.org/reports/global-risk-index-for-ai-enabled-biological-tools/","type":"report"}]}]},
{"date":"2026-01-15","headline":"PDF edition of the handbook available; Handbook citation adds a DOI; BIORISKEVAL tests bio-foundation model safety filtering (Evo 2); Global Risk Index for AI-enabled biological tools (Webster et al., 2025)","entries":[{"category":"milestone","title":"PDF edition of the handbook available","change":"Adds a full PDF edition of the handbook for download, formatted with book-style typography, fonts, and metadata.","why":"Gives readers an offline, printable edition of the full handbook.","chapters":[],"sources":[]},{"category":"milestone","title":"Handbook citation adds a DOI","change":"The Welcome and References pages add a formal citation for the handbook (Tegomoh, 2025, DOI 10.5281/zenodo.18252920).","why":"Gives readers a stable, citable reference for the handbook.","chapters":[["Welcome","/"],["References","/references.html"]],"sources":[{"label":"Tegomoh, 2025","url":"https://doi.org/10.5281/zenodo.18252920","type":"other"}]},{"category":"evidence","title":"BIORISKEVAL tests bio-foundation model safety filtering (Evo 2)","change":"How AI Amplifies Biosecurity Risk adds a case study on Evo 2, a genomic model trained without eukaryotic viral sequences, and BIORISKEVAL, which found fine-tuning restored filtered-virus performance within about 50 training steps.","why":"A 2025 Scale AI/SecureBio preprint tested whether excluding risky training data actually blocks a bio-foundation model from acquiring related capabilities.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Scale AI and SecureBio, 2025, preprint","url":"https://arxiv.org/abs/2510.27629","type":"preprint"},{"label":"Arc Institute, Evo 2 model page","url":"https://arcinstitute.org/tools/evo","type":"other"}]},{"category":"evidence","title":"Global Risk Index for AI-enabled biological tools (Webster et al., 2025)","change":"Policy Frameworks for AI-Bio Convergence and How AI Amplifies Biosecurity Risk cite a CLTR-RAND index of over 1,100 biological tools, with 57 evaluated in depth and 13 flagged 'Red'; 61.5% of Red-flagged tools are fully open source.","why":"Gives readers a cross-tool risk rubric instead of incompatible internal company assessments.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"Webster et al., 2025","url":"https://doi.org/10.71172/wjyw-6dyc","type":"paper"}]}]},
{"date":"2026-01-18","headline":"Red-Teaming AI Systems: Anthropic's ASL-3 activation for Claude Opus 4; LLMs and information hazards: jailbreak vulnerability evidence; Virology Capabilities Test: AI outperforms expert virologists; New case examples in the foundational chapters","entries":[{"category":"developments","title":"Red-Teaming AI Systems: Anthropic's ASL-3 activation for Claude Opus 4","change":"Adds Anthropic's account of activating ASL-3 safeguards for Claude Opus 4 in May 2025 as a precautionary measure after evaluations showed improving CBRN capabilities, without a confirmed threshold breach, plus its Frontier Red Team's continuous testing and OpenAI's pre-deployment evaluation approach.","why":"Gives readers the developer's own account of when enhanced safeguards were triggered, labeled as reported rather than independently confirmed.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Anthropic, 2025, activating ASL-3 report","url":"https://www.anthropic.com/activating-asl3-report","type":"announcement"},{"label":"Anthropic, 2025, nuclear safeguards collaboration","url":"https://red.anthropic.com/2025/nuclear-safeguards/","type":"announcement"},{"label":"Anthropic, 2025, biorisk red-teaming","url":"https://red.anthropic.com/2025/biorisk/","type":"announcement"},{"label":"Anthropic, Frontier Red Team","url":"https://www.anthropic.com/news/strategic-warning-for-ai-risk-progress-and-insights-from-our-frontier-red-team","type":"announcement"},{"label":"OpenAI, updated Preparedness Framework, 2025","url":"https://openai.com/index/updating-our-preparedness-framework/","type":"announcement"}]},{"category":"evidence","title":"LLMs and information hazards: jailbreak vulnerability evidence","change":"Red-Teaming AI Systems cites evidence that large language models respond less stringently to biological information hazards than to other categories of risk, leaving them more exposed to prompt injection and jailbreak attacks.","why":"Gives readers the underlying evidence for why jailbreak defenses need special attention for biological risk content.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"2023 preprint on LLM responses to information hazards","url":"https://arxiv.org/abs/2308.03825","type":"preprint"},{"label":"Promptfoo, how to jailbreak LLMs (blog)","url":"https://www.promptfoo.dev/blog/how-to-jailbreak-llms/","type":"announcement"}]},{"category":"evidence","title":"Virology Capabilities Test: AI outperforms expert virologists","change":"Red-Teaming AI Systems adds the Virology Capabilities Test (VCT), on which OpenAI's o3 model reached 43.8% accuracy on virology troubleshooting questions, ahead of 94% of expert virologists (who averaged 22.1% in their own sub-specialties); OpenAI reports red teamers spent over 1,000 hours testing o3 and o4-mini for biorisk.","why":"Gives readers a concrete measure of AI uplift risk in a specialized biological skill.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Gotting et al., 2025, Virology Capabilities Test, preprint","url":"https://arxiv.org/html/2504.16137v1","type":"preprint"},{"label":"TechCrunch, 2025, OpenAI o3/o4-mini biorisk safeguards","url":"https://techcrunch.com/2025/04/16/openais-latest-ai-models-have-a-new-safeguard-to-prevent-biorisks/","type":"news"}]},{"category":"content","title":"New case examples in the foundational chapters","change":"Pathogens of Concern and The Biological Threat Landscape add case detail on Aum Shinrikyo's toxin research, the 2004 SARS Beijing lab-acquired infections, the 2014 discovery of forgotten CDC smallpox vials, the Soviet Biopreparat program, and the 2001 anthrax letters, plus a citation on underreporting of lab-acquired infections.","why":"Grounds the chapters' historical examples in named, sourced incidents.","chapters":[["Pathogens of Concern","/foundations/pathogens-of-concern.html"],["The Biological Threat Landscape","/foundations/threat-landscape.html"]],"sources":[{"label":"Aum Shinrikyo, Wikipedia","url":"https://en.wikipedia.org/wiki/Aum_Shinrikyo#Biological_weapons","type":"other"},{"label":"The Scientist, 2014, SARS escaped Beijing lab twice","url":"https://www.the-scientist.com/news-analysis/sars-escaped-beijing-lab-twice-50137","type":"news"},{"label":"NPR, 2014, smallpox vials found in lab storeroom","url":"https://www.npr.org/2014/07/08/329884145/in-a-lab-store-room-an-unsettling-surprise-lost-vials-of-smallpox","type":"news"},{"label":"Federation of American Scientists, Soviet Biopreparat","url":"https://nuke.fas.org/guide/russia/agency/bw.htm","type":"other"},{"label":"FBI, Amerithrax investigation","url":"https://www.fbi.gov/history/famous-cases/amerithrax-or-anthrax-investigation","type":"policy"},{"label":"PMC paper on underreporting of laboratory-acquired infections","url":"https://pmc.ncbi.nlm.nih.gov/articles/PMC7348994/","type":"paper"}]}]},
{"date":"2026-01-19","headline":"'Common Questions' FAQ added to four chapters; DNA synthesis cost trend cited (Carlson, 2009)","entries":[{"category":"content","title":"'Common Questions' FAQ added to four chapters","change":"Adds a Common Questions section to Medical Countermeasures and Biodefense, Dual-Use Research of Concern (DURC), Laboratory Biosafety and Biosecurity, and Outbreak Detection and Surveillance, covering the Strategic National Stockpile, Operation Warp Speed, the 2011-2012 H5N1 ferret controversy, BSL containment levels, and surveillance methods.","why":"Gives readers quick answers to common questions instead of requiring a full chapter read.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[]},{"category":"evidence","title":"DNA synthesis cost trend cited (Carlson, 2009)","change":"Synthetic Biology and Democratization cites Carlson (2009) for the finding that gene synthesis prices have historically halved roughly every 15 months, a pace the chapter compares to Moore's Law.","why":"Gives readers a citable source for the chapter's cost-decline claim.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"]],"sources":[{"label":"Carlson, 2009","url":"https://www.nature.com/articles/nbt1209-1091","type":"paper"}]}]},
{"date":"2026-01-22","headline":"'Paraphrase Project': AI-redesigned toxins evaded DNA screening; NIST AI 800-1: managing misuse risk for foundation models; Value alignment framework adapted from clinical AI (VIM); PREVENT Pandemics Act (2022) added to policy coverage","entries":[{"category":"evidence","title":"'Paraphrase Project': AI-redesigned toxins evaded DNA screening","change":"AI-Enabled Pathogen Design and DNA Synthesis Screening describe the Paraphrase Project (Wittmann et al., Science, October 2025), which used the AI tool EvoDiff to redesign known toxins that evaded BLAST-based screening; detection was as low as 23% before patching and reached 97% for the highest-risk sequences afterward, with about 3% still escaping detection.","why":"Documents a demonstrated screening bypass and the industry response that closed most of the gap.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Wittmann et al., Science, 2025","url":"https://www.science.org/doi/10.1126/science.adu8578","type":"paper"}]},{"category":"policy","title":"NIST AI 800-1: managing misuse risk for foundation models","change":"Policy Frameworks for AI-Bio Convergence adds a section on NIST AI 800-1, the U.S. framework for identifying, measuring, and mitigating foundation model misuse risk, including its Second Public Draft appendix on chemical and biological threats; Red-Teaming AI Systems and LLMs and Information Hazards add shorter references to the same framework and the renamed UK and U.S. AI safety institutes.","why":"Gives readers the primary U.S. government framework for foundation model misuse risk, including an independent assessment of its biological-risk coverage.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"NIST AI 800-1, Second Public Draft","url":"https://doi.org/10.6028/NIST.AI.800-1.2pd","type":"policy"},{"label":"NIST, updated guidelines for dual-use foundation models, January 2025","url":"https://www.nist.gov/news-events/news/2025/01/updated-guidelines-managing-misuse-risk-dual-use-foundation-models","type":"policy"},{"label":"NIST AI 800-1, Second Public Draft (PDF)","url":"https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.800-1.ipd2.pdf","type":"policy"},{"label":"Johns Hopkins Center for Health Security, AI x Bio","url":"https://centerforhealthsecurity.org/our-work/aixbio","type":"report"},{"label":"Johns Hopkins Center for Health Security, comment on NIST AI 800-1","url":"https://centerforhealthsecurity.org/sites/default/files/2025-04/NIST-AI-800-1-RFC-3.15.25-Johns-Hopkins-Center-for-Health-Security_edited.pdf","type":"report"}]},{"category":"evidence","title":"Value alignment framework adapted from clinical AI (VIM)","change":"Policy Frameworks for AI-Bio Convergence adds a section applying the RAISE consortium's 'Values In the Model' (VIM) framework, developed for clinical AI, to biosecurity, notes the MedLog protocol as a possible adaptation for logging AI interactions, and lists RAISE in the resources appendix.","why":"Offers a transparency framework from another domain that biosecurity governance does not yet have.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"Goldberg et al., NEJM AI, 2026","url":"https://doi.org/10.1056/AIp2501266","type":"paper"},{"label":"Noori et al., 2025, preprint","url":"https://arxiv.org/abs/2510.04033","type":"preprint"},{"label":"Harvard Medical School Center for Bioethics, RAISE symposium","url":"https://bioethics.hms.harvard.edu/events/2nd-international-symposium-responsible-ai-social-ethical-healthcare-raise","type":"other"}]},{"category":"policy","title":"PREVENT Pandemics Act (2022) added to policy coverage","change":"Policy Frameworks for AI-Bio Convergence and Dual-Use Research of Concern (DURC) add coverage of the PREVENT Pandemics Act (P.L. 117-328, enacted December 2022), which strengthened NSABB's authority, mandated a federal strategy for high-containment labs, and created the White House Office of Pandemic Preparedness and Response Policy, while leaving AI-specific oversight gaps.","why":"Gives readers the statutory basis for NSABB's current authority and shows where existing law does not address AI.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"sources":[{"label":"PREVENT Pandemics Act, S.3799, 117th Congress","url":"https://www.congress.gov/bill/117th-congress/senate-bill/3799","type":"policy"},{"label":"Congressional Research Service, R47649","url":"https://www.congress.gov/crs-product/R47649","type":"policy"},{"label":"White House, Office of Pandemic Preparedness and Response Policy, 2023","url":"https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-white-house-launches-office-of-pandemic-preparedness-and-response-policy/","type":"policy"}]}]},
{"date":"2026-01-24","headline":"LLM vs. biological design tool risk framing clarified; Anthropic's precautionary ASL-3 activation","entries":[{"category":"revision","title":"LLM vs. biological design tool risk framing clarified","change":"How AI Amplifies Biosecurity Risk clarifies the distinction between LLMs and biological design tools as distinct risk categories, and adds a note that the widely cited MIT classroom exercise (Soice et al., 2023, preprint) was a demonstration, not an operational feasibility test.","why":"Prevents readers from mistaking a classroom exercise for evidence of real-world feasibility.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[]},{"category":"revision","title":"Anthropic's precautionary ASL-3 activation","change":"Red-Teaming AI Systems describes Anthropic's May 2025 activation of ASL-3 safeguards for Claude Opus 4 as a precautionary measure rather than a confirmed threshold breach, notes RAND's finding that current LLMs do not dramatically increase bioweapons risk beyond internet search, and references Anthropic's Constitutional Classifiers++.","why":"Distinguishes precautionary safeguards from confirmed threshold breaches and explains why current uplift findings may not hold as capabilities advance.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[]}]},
{"date":"2026-01-25","headline":"Biopreparat program scale added to Case Studies","entries":[{"category":"revision","title":"Biopreparat program scale added to Case Studies","change":"Case Studies adds that the Soviet Biopreparat program employed over 30,000 personnel at its peak and that the full Soviet biological weapons complex, including military institutes, exceeded 60,000 people.","why":"Gives readers the scale of the covert Soviet biological weapons program.","chapters":[["Case Studies in Biosecurity Failures and Successes","/governance/case-studies.html"]],"sources":[]}]},
{"date":"2026-01-27","headline":"DNA Synthesis Screening adds reported turnaround times and fees","entries":[{"category":"revision","title":"DNA Synthesis Screening adds reported turnaround times and fees","change":"DNA Synthesis Screening notes that flagged orders typically take synthesis companies 2 to 5 business days to review and may carry modest verification fees, while routine orders see no added delay or cost.","why":"Answers a practical question researchers raise about the cost and speed of the screening process.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[]}]},
{"date":"2026-02-01","headline":"New Cloud Labs section on autonomous AI agents as an attack surface","entries":[{"category":"evidence","title":"Agent skill supply chain found to carry widespread vulnerabilities","change":"Cloud Labs and Automated Biology adds a section on the AI agent 'skill' supply chain, citing a January 2026 study finding 26.1% of 42,447 agent skills contained a security vulnerability and 5.2% showed signs of malicious intent, plus a reported demonstration of a backdoored skill topping download charts in seven countries.","why":"Gives readers evidence that the plugin ecosystem for autonomous lab agents already carries measurable security risk.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"Liu et al., 2026, preprint","url":"https://arxiv.org/abs/2601.10338","type":"preprint"},{"label":"The Stack, January 2026","url":"https://www.thestack.technology/a-backdoor-was-the-most-downloaded-skill-for-viral-clawdbot-moltbot-and-why-that-matters/","type":"news"}]},{"category":"developments","title":"Galileo AI reports rapid poisoning risk in multi-agent systems","change":"Cloud Labs and Automated Biology cites Galileo AI's reported finding that in simulated multi-agent systems, a single compromised agent could poison 87% of downstream decision-making within four hours through cascading influence on shared memory and task delegation.","why":"Adds a reported estimate of how quickly a single compromised agent could corrupt outputs across a multi-agent lab workflow.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"Galileo AI, 2025","url":"https://galileo.ai/blog/malicious-behavior-in-multi-agent-systems","type":"announcement"}]},{"category":"developments","title":"OWASP Top 10 for Agentic Applications cited for cloud lab risks","change":"Cloud Labs and Automated Biology cites the OWASP Top 10 for Agentic Applications, released December 2025, for risks directly relevant to cloud lab security.","why":"Gives readers an industry framework for identifying agentic security risks relevant to cloud labs.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"OWASP, Top 10 for Agentic Applications, 2025","url":"https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/","type":"report"}]},{"category":"evidence","title":"Royal Society Open Science review of self-driving lab safety","change":"Cloud Labs and Automated Biology cites a July 2025 review in Royal Society Open Science examining the safety and security implications of self-driving laboratories, where AI systems control physical laboratory equipment directly.","why":"Gives readers a peer-reviewed assessment of the added risks when AI directly operates physical lab equipment.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"Royal Society Open Science review, 2025","url":"https://royalsocietypublishing.org/doi/10.1098/rsos.250646","type":"paper"}]},{"category":"policy","title":"NIST requests input on AI agent security","change":"Cloud Labs and Automated Biology notes that NIST's Center for AI Standards and Innovation issued a January 2026 Request for Information on AI agent security, citing risks of hijacking, backdoor attacks, and other exploits, with comments due March 9.","why":"Signals that federal standards attention now extends to the AI agent risks the chapter describes.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"NIST CAISI, Request for Information on AI agent security, 2026","url":"https://www.federalregister.gov/documents/2026/01/08/2026-00206/request-for-information-regarding-security-considerations-for-artificial-intelligence-agents","type":"policy"}]}]},
{"date":"2026-02-03","headline":"New chapter: Autonomous AI Agents in Laboratory Contexts","entries":[{"category":"content","title":"New chapter: Autonomous AI Agents in Laboratory Contexts","change":"Autonomous AI Agents treats the laboratory attack surface, the skill supply chain, multi-agent coordination, and self-driving laboratory risks.","why":"Gives readers dedicated treatment of AI agents that can now operate laboratory infrastructure without human oversight.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Executive Summary","/executive-summary.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"]],"sources":[]},{"category":"developments","title":"OpenClaw's rise and the Moltbook agent social network","change":"Autonomous AI Agents cites reporting that OpenClaw, an open-source autonomous agent framework, passed 100,000 GitHub stars, and that its users created Moltbook, an agent-only social network that grew to over 1.5 million registered agents by January 2026.","why":"Shows readers how quickly open-source agent frameworks and agent-to-agent platforms are scaling outside major AI labs.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"OpenClaw","url":"https://openclaw.ai/","type":"other"},{"label":"Moltbook","url":"https://moltbook.com","type":"other"},{"label":"CNBC, February 2026","url":"https://www.cnbc.com/2026/02/02/openclaw-open-source-ai-agent-rise-controversy-clawdbot-moltbot-moltbook.html","type":"news"}]}]},
{"date":"2026-02-05","headline":"New appendix: Course Syllabus template; Executive Summary adds autonomous AI agent risk and cloud lab governance; OpenAI reports GPT-5 collaboration on wet-lab experiment design; Cloud Labs clarifies AI wet-lab capability as theoretical, not demonstrated","entries":[{"category":"content","title":"New appendix: Course Syllabus template","change":"Adds a Course Syllabus appendix with a 12-week (and condensed 6-week) graduate course structure built around the handbook, plus supplementary reading lists and instructor notes.","why":"Gives instructors a ready-to-adapt structure for teaching biosecurity and AI convergence in graduate programs.","chapters":[["Course Syllabus Template","/appendices/course-syllabus.html"],["Preface","/preface.html"]],"sources":[]},{"category":"revision","title":"Executive Summary adds autonomous AI agent risk and cloud lab governance","change":"Executive Summary adds a section on autonomous AI agents as a new attack surface, citing the 26.1% agent-skill vulnerability rate and 87% downstream-compromise propagation figures, and recommends API access controls, audit logging, and middleware security standards for cloud laboratories.","why":"Raises the priority of AI agent risk in the handbook's top-level risk assessment and gives institutions concrete governance steps.","chapters":[["Executive Summary","/executive-summary.html"]],"sources":[]},{"category":"developments","title":"OpenAI reports GPT-5 collaboration on wet-lab experiment design","change":"How AI Amplifies Biosecurity Risk describes a December 2025 demonstration in which OpenAI's GPT-5 iterated on wet-lab experimental design and troubleshooting with biosecurity firm Red Queen Bio, reportedly achieving a 79-fold improvement in molecular cloning efficiency while human scientists executed the physical lab work under biosecurity safeguards.","why":"Marks a reported step from AI giving lab information toward AI directing lab experiments, though humans still performed the physical work.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"OpenAI, accelerating biological research in the wet lab","url":"https://openai.com/index/accelerating-biological-research-in-the-wet-lab/","type":"announcement"}]},{"category":"revision","title":"Cloud Labs clarifies AI wet-lab capability as theoretical, not demonstrated","change":"Cloud Labs and Automated Biology clarifies that AI agents conducting chemistry experiments are already demonstrated, but full autonomous AI-driven biological experimentation through cloud laboratories remains theoretical, with no published demonstration to date.","why":"Keeps the chapter's claims about AI-run biology precisely calibrated between what has been shown and what remains hypothetical.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[]}]},
{"date":"2026-02-12","headline":"WMDP-Bio, Virology Capabilities Test, and BIORISKEVAL added as evaluation benchmarks; Further Reading adds AI safety course and textbook","entries":[{"category":"evidence","title":"WMDP-Bio, Virology Capabilities Test, and BIORISKEVAL added as evaluation benchmarks","change":"How AI Amplifies Biosecurity Risk adds the Center for AI Safety's WMDP-Bio benchmark (1,273 biosecurity questions) and its unlearning-method evaluations. LLMs and Information Hazards compares WMDP-Bio with the Virology Capabilities Test (322 virology questions on which models reportedly outperform 94% of expert virologists) and the BIORISKEVAL framework.","why":"Gives readers named, citable benchmarks for comparing how models are evaluated for biosecurity risk.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Li, Mazeika, Hendrycks et al., 2024","url":"https://arxiv.org/abs/2403.03218","type":"preprint"},{"label":"Virology Capabilities Test benchmark","url":"https://arxiv.org/abs/2504.16137","type":"preprint"}]},{"category":"content","title":"Further Reading adds AI safety course and textbook","change":"Further Reading includes the Center for AI Safety's free Introduction to AI Safety, Ethics, and Society course and its companion textbook by Hendrycks (2024).","why":"Readers who need grounding in catastrophic risk, including biological threats, can work through an AI safety course and its companion textbook at no cost.","chapters":[["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"],["Preface","/preface.html"]],"sources":[{"label":"Hendrycks, Introduction to AI Safety, Ethics, and Society, 2024","url":"https://doi.org/10.1201/9781003530336","type":"paper"},{"label":"Center for AI Safety, Introduction to AI Safety, Ethics, and Society (virtual course)","url":"https://www.aisafetybook.com/virtual-course","type":"other"}]}]},
{"date":"2026-02-18","headline":"Biosecurity Modernization and Innovation Act of 2026 added to DNA screening coverage","entries":[{"category":"policy","title":"Biosecurity Modernization and Innovation Act of 2026 added to DNA screening coverage","change":"DNA Synthesis Screening, the Executive Summary, and Policy Frameworks for AI-Bio Convergence add the bipartisan Biosecurity Modernization and Innovation Act of 2026 (Cotton/Klobuchar), which would direct the Commerce Department to mandate synthesis screening, a dangerous-sequence registry, and a NIST governance sandbox. The chapter notes the bill remains before the Senate.","why":"Tracks the first bipartisan legislative effort to mandate DNA synthesis screening nationwide.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Executive Summary","/executive-summary.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Cotton, Klobuchar, press release on Biosecurity Modernization and Innovation Act, 2026","url":"https://www.cotton.senate.gov/news/press-releases/cotton-klobuchar-introduce-bill-to-establish-federal-biotech-security-framework","type":"policy"},{"label":"NTI, endorsement of the Biosecurity Modernization and Innovation Act of 2026","url":"https://www.nti.org/news/nti-endorses-biosecurity-modernization-and-innovation-act-of-2026/","type":"report"},{"label":"FAS, on the Biosecurity Modernization and Innovation Act of 2026","url":"https://fas.org/publication/biosecurity-modernization-and-innovation-act-of-2026/","type":"report"}]}]},
{"date":"2026-02-20","headline":"New Biosecurity Data Levels framework added across AI-bio chapters","entries":[{"category":"policy","title":"AI risk chapter updated with DNA screening bills status","change":"How AI Amplifies Biosecurity Risk updates its critical chokepoint discussion to note that the OSTP framework revision deadline passed without a replacement, and that Congress has introduced H.R. 3029 (NIST screening standards) and the Biosecurity Modernization and Innovation Act (S. 3741, mandatory screening through the Commerce Department).","why":"Keeps the chapter's regulatory status current with pending congressional action on synthesis screening.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"H.R. 3029 bill text","url":"https://www.congress.gov/bill/119th-congress/house-bill/3029/text","type":"policy"},{"label":"Cotton, Klobuchar, press release on Biosecurity Modernization and Innovation Act, 2026","url":"https://www.cotton.senate.gov/news/press-releases/cotton-klobuchar-introduce-bill-to-establish-federal-biotech-security-framework","type":"policy"}]},{"category":"evidence","title":"Biosecurity Data Levels framework added across AI pathogen design coverage","change":"AI-Enabled Pathogen Design adds a five-tier Biosecurity Data Level (BDL-0 through BDL-4) framework for classifying pathogen datasets by their contribution to AI capabilities of concern, proposed by Bloomfield et al. (2026) and endorsed by more than 100 researchers at the Asilomar Conference. LLMs and Information Hazards, Policy Frameworks for AI-Bio Convergence, the Glossary, and the Executive Summary add matching coverage.","why":"Gives readers a peer-reviewed proposal for governing pathogen training data before models are built.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["Executive Summary","/executive-summary.html"]],"sources":[{"label":"Bloomfield et al., 2026","url":"https://doi.org/10.1126/science.aeb2689","type":"paper"}]}]},
{"date":"2026-02-22","headline":"Biosecurity Modernization Act endorser list expanded with named organizations","entries":[{"category":"revision","title":"Biosecurity Modernization Act endorser list expanded with named organizations","change":"DNA Synthesis Screening and Policy Frameworks for AI-Bio Convergence replace a general reference to 'NTI, FAS, and major synthesis providers' with a full list of endorsers: the National Security Commission on Emerging Biotechnology, the Engineering Biology Research Consortium, Twist Bioscience, Aclid, IDT, and Ginkgo Bioworks.","why":"Shows readers exactly which organizations and companies back the pending legislation.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[]}]},
{"date":"2026-02-23","headline":"Inference-time PPI filters shown to miss SARS-CoV-2 binding mutants","entries":[{"category":"evidence","title":"Inference-time PPI filters shown to miss SARS-CoV-2 binding mutants","change":"AI-Enabled Pathogen Design adds a new section describing a NeurIPS Biosecurity Safeguards workshop study finding that three protein-protein interaction prediction tools proposed as inference-time biosafety filters, AlphaFold 3, AF3Complex, and SpatialPPIv2, detected none of four experimentally validated SARS-CoV-2 binding mutants (Feldman and Feldman, 2025). DNA Synthesis Screening and the chapter summaries page add matching notes on the finding.","why":"Shows readers that a proposed AI-design safety layer failed against a well-studied virus in controlled testing.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"]],"sources":[{"label":"Feldman and Feldman, 2025, workshop paper","url":"https://openreview.net/forum?id=NAr5wbHghS","type":"paper"}]}]},
{"date":"2026-02-25","headline":"RAND risk-scoring tool and DNA screening red-team test added","entries":[{"category":"evidence","title":"RAND risk-scoring tool and DNA synthesis red-team test added","change":"DNA Synthesis Screening adds a red-team finding that 36 of 38 commercial DNA providers shipped fragments sufficient to reconstruct the 1918 influenza genome to a fake organization, with only one provider flagging it (Edison et al., 2026). Policy Frameworks for AI-Bio Convergence and Further Reading add a RAND scoring method pairing viral risk with actor capability (Williams et al., 2026).","why":"Shows readers a real-world gap in DNA synthesis screening and a proposed method for scoring biological AI risk before deployment.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"Edison et al., 2026","url":"https://doi.org/10.1038/s41467-025-67955-3","type":"paper"},{"label":"Williams et al., RAND, 2026","url":"https://www.rand.org/pubs/research_reports/RRA4490-1.html","type":"report"}]}]},
{"date":"2026-03-18","headline":"Anthropic's RSP rollback and a DNA-synthesis-screening red-team test","entries":[{"category":"developments","title":"Anthropic survey on public perception of AI risks cited","change":"How AI Amplifies Biosecurity Risk cites Anthropic's survey of 80,508 users across 159 countries, in which 13% named malicious use, including bioweapons, a primary concern, ranking eighth of thirteen categories behind unreliability and job displacement.","why":"Contrasts public risk perception with expert assessment of biosecurity as a critical frontier risk.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Anthropic, 2025","url":"https://www.anthropic.com/features/81k-interviews","type":"announcement"}]},{"category":"evidence","title":"First RCT finds no measurable uplift from LLM assistance in lab tasks","change":"LLMs and Information Hazards adds Hong et al.'s pre-registered, investigator-blinded RCT (n=153) testing whether LLM assistance improves novice completion of a viral reverse-genetics workflow, finding comparable completion rates (5.2% vs 6.6%, P=0.759) though modestly higher intermediate-step progress.","why":"The first controlled physical-lab test of LLM uplift, addressing a gap left by prior evaluations that measured only knowledge access or planning quality.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Hong et al., 2026, preprint","url":"https://arxiv.org/abs/2602.16703","type":"preprint"}]},{"category":"developments","title":"Anthropic's Responsible Scaling Policy rollback (February 2026)","change":"Red-Teaming AI Systems and Policy Frameworks for AI-Bio Convergence add a case study on Anthropic's February 2026 Responsible Scaling Policy revision, which replaces its categorical commitment to halt training when safeguards lag capabilities with a conditional promise to delay only if Anthropic judges itself the industry leader and catastrophic risk to be high.","why":"A frontier developer weakened a voluntary safety commitment that the handbook's governance discussion had relied on as an industry-leading example.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Anthropic, Responsible Scaling Policy v3.0, 2026","url":"https://anthropic.com/responsible-scaling-policy/rsp-v3-0","type":"announcement"},{"label":"Heiligenstein, TIME, 2026","url":"https://time.com/7380854/exclusive-anthropic-drops-flagship-safety-pledge/","type":"news"},{"label":"CNN, 2026","url":"https://www.cnn.com/2026/02/25/tech/anthropic-safety-policy-change","type":"news"}]},{"category":"evidence","title":"DNA synthesis screening red-team test finds widespread gaps","change":"DNA Synthesis Screening cites an FBI-overseen red-teaming experiment in which 36 of 38 DNA synthesis providers, including 12 of 13 IGSC members, shipped 1918 influenza fragments without detection.","why":"Demonstrates that short, often-unscreened oligonucleotide fragments can evade current synthesis screening in practice.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Edison, Toner, and Esvelt, Nature Communications, 2026","url":"https://www.nature.com/articles/s41467-025-67955-3","type":"paper"}]},{"category":"policy","title":"GAO finds HHS gain-of-function oversight lacks transparency","change":"Gain-of-Function Research cites a January 2026 GAO report finding that HHS does not consistently disclose how it assesses and mitigates risks in high-risk pathogen research, and recommending regular public reporting on review outcomes.","why":"An independent government investigation validates the chapter's existing critique of opaque gain-of-function review.","chapters":[["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"]],"sources":[{"label":"GAO, GAO-26-107348, 2026","url":"https://www.gao.gov/products/gao-26-107348","type":"policy"}]},{"category":"policy","title":"NIH Biosafety Modernization Initiative added","change":"Laboratory Biosafety and Biosecurity adds NIH's 2025 Biosafety Modernization Initiative, which proposes shifting to a risk-based biosafety framework, elevating Institutional Biosafety Committees to IRB-like oversight authority, and requires public posting of IBC meeting minutes for sessions held on or after June 1, 2025.","why":"Documents an active NIH effort to overhaul the oversight framework the chapter describes.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"NIH Office of Science Policy, biosafety and biosecurity policy","url":"https://osp.od.nih.gov/policies/biosafety-and-biosecurity-policy/","type":"policy"}]},{"category":"evidence","title":"Airport wastewater surveillance modeling study added","change":"Outbreak Detection and Surveillance adds a modeling study estimating that airport aircraft-wastewater sampling could improve first-detection times by 12.5 to 37.7 days over a range of scenarios, and by 22.0 to 25.6 days for a SARS-CoV-2-like pathogen compared with ICU-based genomic monitoring.","why":"Offers a scalable early-detection tool that operationalizes the IHR's existing points-of-entry surveillance mandate.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"Bhatt et al., 2026, preprint","url":"https://www.medrxiv.org/content/10.64898/2026.02.27.26347250v1","type":"preprint"}]},{"category":"revision","title":"H5N1 dairy cattle case study updated with current case counts","change":"Case Studies updates the ongoing H5N1 dairy cattle outbreak case study to CDC's count of 71 total U.S. human A(H5) cases since February 2024, and notes CDC's shift to monthly targeted-surveillance reporting in July 2025, directing readers to USDA for animal detection totals.","why":"Keeps the chapter's account of an unresolved, evolving outbreak current.","chapters":[["Case Studies in Biosecurity Failures and Successes","/governance/case-studies.html"]],"sources":[]},{"category":"policy","title":"America's Living Library Act of 2026 introduced","change":"Policy Frameworks for AI-Bio Convergence adds bipartisan, bicameral legislation directing the Department of the Interior to collect and sequence genomic information from U.S. National Parks, creating standardized AI-ready biological datasets for biotech innovation.","why":"Tracks a new legislative proposal for public biological data infrastructure relevant to AI-bio governance.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Padilla, Young, Bice, and Khanna, America's Living Library Act of 2026","url":"https://www.padilla.senate.gov/newsroom/press-releases/padilla-young-bice-khanna-introduce-americas-living-library-act-to-strengthen-u-s-biotech-innovation/","type":"policy"}]},{"category":"evidence","title":"NTI AIxBio Horizon Scan on near-term capability trends added","change":"Policy Frameworks for AI-Bio Convergence cites NTI's biannual AIxBio Horizon Scan, whose March 2026 analysis identifies LLM integration with biological research tools, automated laboratory platforms, and closed-loop discovery systems as trends requiring governance attention within 18 months.","why":"Gives the chapter a forward-looking forecast of which AI-bio capability trends need near-term governance attention.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"NTI, AIxBio Horizon Scan, Winter 2025-2026","url":"https://www.nti.org/analysis/articles/aixbio-horizon-scan-winter-2025-2026/","type":"report"},{"label":"NTI, AIxBio safety and security analysis, March 2026","url":"https://www.nti.org/risky-business/a-critical-moment-for-aixbio-safety-and-security/","type":"report"}]}]},
{"date":"2026-03-28","headline":"GAO international biosafety and biosecurity comparison added; GISAID sequence count updated and data feed restriction noted; Wastewater surveillance design framework added; WHO Pandemic Agreement PABS annex negotiations documented","entries":[{"category":"policy","title":"GAO international biosafety and biosecurity comparison added","change":"Laboratory Biosafety and Biosecurity adds a 2026 GAO comparison of the United States and selected G20 members across 10 program components, finding 9 of 10 countries had guidance comparable to at least one U.S. component but common gaps in biosecurity-specific language and inventory management.","why":"Gives the chapter's containment-focused discussion a comparative view of what complete laboratory governance requires beyond containment rules.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"GAO, GAO-26-107338, 2026","url":"https://www.gao.gov/products/gao-26-107338","type":"policy"}]},{"category":"developments","title":"GISAID sequence count updated and data feed restriction noted","change":"Outbreak Detection and Surveillance updates GISAID's shared SARS-CoV-2 sequence count to over 16 million and adds that GISAID ended the flat-file data feed that Nextstrain and other downstream tools had relied on since February 2020, restricting automated access.","why":"Keeps the sequence count current and documents a change that disrupted real-time genomic visualization tools built on GISAID data.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"Nextstrain, 2025","url":"https://nextstrain.org/blog/2025-11-06-gisaid-based-ncov-analyses","type":"other"}]},{"category":"evidence","title":"Wastewater surveillance design framework added","change":"Outbreak Detection and Surveillance adds Nikiforuk et al.'s framework for wastewater surveillance design, which argues programs should explicitly define six choices: pathogen suitability, surveillance objective, sampling strategy, molecular assays, data transformation, and analysis.","why":"Gives the chapter a structured framework for designing wastewater surveillance programs rather than treating the signal as uniformly interpretable.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"Nikiforuk et al., 2026","url":"https://doi.org/10.1038/s41545-026-00570-x","type":"paper"}]},{"category":"policy","title":"WHO Pandemic Agreement PABS annex negotiations documented","change":"Global Surveillance Equity adds context on the WHO Pandemic Agreement's Pathogen Access and Benefit Sharing annex, noting that Member States held negotiating rounds in 2026 to finalize rapid pathogen-sharing and benefit-access rules ahead of the 79th World Health Assembly, extending the 2011 PIP Framework's benefit-sharing logic beyond influenza.","why":"Tracks the clearest current attempt to close the benefit-sharing gap the chapter identifies in the influenza-only PIP Framework.","chapters":[["Global Surveillance Equity: LMIC Capacity as Collective Security","/governance/surveillance-equity.html"]],"sources":[{"label":"WHO, Pandemic Agreement PABS annex negotiations, 2026","url":"https://www.who.int/news/item/17-02-2026-global-commitment-on-display-as-countries-negotiate-key-annex-to-the-pandemic-agreement","type":"policy"},{"label":"WHO Director-General, remarks, 2026","url":"https://www.who.int/news-room/speeches/item/who-director-general-s-opening-remarks-at-the-sixth-intergovernmental-working-group-(igwg)-on-the-who-pandemic-agreement-23-march-2026","type":"policy"}]}]},
{"date":"2026-04-01","headline":"UK Pandemic Preparedness Strategy (2026) added; AI-enabled biothreat safeguards mapped across the development chain; WHA77.7 laboratory biosafety resolution added; Proposals for standardized BSL-3/4 reporting added; Health Security landscape review of AI biodefense applications added","entries":[{"category":"policy","title":"UK Pandemic Preparedness Strategy (2026) added","change":"Policy Frameworks for AI-Bio Convergence adds the UK's March 2026 Pandemic Preparedness Strategy, an all-hazards framework replacing older influenza-focused planning that spans respiratory, blood-borne, and vector-borne routes under a One Health lens, alongside the associated All Pandemic Hazards Bill.","why":"Documents a new national pandemic-preparedness framework relevant to the chapter's policy coverage.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"UK Department of Health and Social Care, Pandemic Preparedness Strategy, 2026","url":"https://www.gov.uk/government/publications/pandemic-preparedness-strategy-building-our-capabilities/introduction-and-principles-uk-wide","type":"policy"},{"label":"UK Department of Health and Social Care, All Pandemic Hazards Bill, 2026","url":"https://www.gov.uk/government/news/1-billion-invested-in-health-protection-as-new-pandemic-strategy-published","type":"policy"}]},{"category":"evidence","title":"AI-enabled biothreat safeguards mapped across the development chain","change":"Policy Frameworks for AI-Bio Convergence adds a synthesis proposing safeguards organized around the full chain of AI-enabled biothreat creation, from upstream model development and access controls to downstream synthesis screening and response infrastructure.","why":"Gives the chapter a framework for mapping mitigations to specific failure points rather than relying on a single safeguard.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Dettman et al., 2026","url":"https://doi.org/10.1002/bit.70132","type":"paper"}]},{"category":"policy","title":"WHA77.7 laboratory biosafety resolution added","change":"Laboratory Biosafety and Biosecurity adds the World Health Assembly's June 2024 resolution WHA77.7, which calls on Member States to strengthen national laboratory biosafety and biosecurity strategies and extends biological risk management to genetic engineering, synthetic biology, and related data, not just physical samples.","why":"Documents an international resolution extending the chapter's risk-management scope beyond physical samples.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"World Health Assembly, WHA77.7, 2024","url":"https://apps.who.int/gb/ebwha/pdf_files/WHA77/A77_R7-en.pdf","type":"policy"}]},{"category":"evidence","title":"Proposals for standardized BSL-3/4 reporting added","change":"Laboratory Biosafety and Biosecurity adds a 2025 Nature Methods proposal for the MIHCLE reporting standard and a 2026 PLOS Biology perspective, both arguing that funders and journals should require standardized reporting for BSL-3 and BSL-4 experiments.","why":"Treats structured reporting and gatekeeping as biosafety and biosecurity controls rather than publication hygiene.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"Nature Methods, MIHCLE reporting standard proposal, 2025","url":"https://www.nature.com/articles/s41592-025-02862-5","type":"paper"},{"label":"PLOS Biology, perspective, 2026","url":"https://doi.org/10.1371/journal.pbio.3003691","type":"paper"}]},{"category":"evidence","title":"Health Security landscape review of AI biodefense applications added","change":"AI for Biosecurity Defense adds a 2026 Health Security landscape review of AI applications across biosecurity, biodefense, and pandemic preparedness, concluding the most actionable near-term gains are in outbreak prediction, resource allocation, countermeasure acceleration, and laboratory workflow support.","why":"Supports the chapter's case for investing in detection and response speed over speculative future capabilities.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"Attal-Juncqua et al., 2026","url":"https://doi.org/10.1177/23265094261418815","type":"paper"}]}]},
{"date":"2026-04-08","headline":"Practitioner study finds DURC/PEPP implementation barriers; Review finds international biosafety governance fragmented across institutions","entries":[{"category":"evidence","title":"Practitioner study finds DURC/PEPP implementation barriers","change":"Policy Frameworks for AI-Bio Convergence adds a grounded-theory study of 45 U.S. biosafety officers, compliance professionals, and researchers, who supported DURC/PEPP's oversight goals but identified ambiguous definitions, limited legal scope, missing federal guidance, and unfunded administrative burdens as major implementation barriers.","why":"The barriers practitioners identified in the 2024 framework will face any successor policy.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Gillum et al., Politics and the Life Sciences, 2025","url":"https://doi.org/10.1017/pls.2025.10014","type":"paper"}]},{"category":"evidence","title":"Review finds international biosafety governance fragmented across institutions","change":"Policy Frameworks for AI-Bio Convergence adds a review arguing that biological capability is now widely distributed while institutional oversight remains fragmented across health, security, and environmental bodies, leaving no single forum equipped to govern cloud laboratories, AI-enabled design tools, and cross-border data flows coherently.","why":"Extends the chapter's governance-gap discussion from domestic DURC/PEPP implementation to the broader international governance architecture for emerging biological technologies.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Lentzos and Bowsher, Cold Spring Harbor Perspectives in Medicine, 2026, preprint","url":"https://doi.org/10.1101/cshperspect.a041627","type":"preprint"}]}]},
{"date":"2026-04-12","headline":"Autonomous agents chapter expanded with 2026 security research","entries":[{"category":"evidence","title":"NIST red-teaming competition finds attacks succeed against every frontier agent","change":"Autonomous AI Agents adds NIST CAISI's March 2026 agent-security red-teaming competition, in which more than 400 participants made over 250,000 attack attempts against 13 frontier models, with at least one successful attack against every target, including scenarios targeting multi-agent tool use and coordination.","why":"Provides large-scale empirical evidence that current frontier agents are vulnerable to adversarial manipulation.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"NIST CAISI, agent-security red-teaming competition, 2026","url":"https://www.nist.gov/blogs/caisi-research-blog/insights-ai-agent-security-large-scale-red-teaming-competition","type":"policy"},{"label":"Debenedetti et al., 2024, preprint","url":"https://arxiv.org/abs/2406.13352","type":"preprint"}]},{"category":"policy","title":"NIST launches AI Agent Standards Initiative","change":"Autonomous AI Agents adds CAISI's February 2026 AI Agent Standards Initiative, organized around a trusted, interoperable, and secure agent ecosystem, with governance themes covering interoperable agent protocols, agent identity infrastructure, and authorization models for human-agent and multi-agent interactions.","why":"Tracks a new federal standards effort directly relevant to securing laboratory-connected agent deployments.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"NIST CAISI, AI Agent Standards Initiative, 2026","url":"https://www.nist.gov/caisi/ai-agent-standards-initiative","type":"policy"}]},{"category":"evidence","title":"Security frameworks catalog multi-agent and agent attack surfaces","change":"Autonomous AI Agents adds two March 2026 preprints arguing multi-agent systems are a distinct security domain: one scoring 16 security frameworks against 193 threat items around delegated tool authority and shared memory, the other cataloging attack surfaces especially relevant to laboratory infrastructure.","why":"Gives the chapter a systematic taxonomy of agent-specific attack surfaces beyond single-model risk.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Nguyen et al., Security Considerations for Multi-agent Systems, 2026, preprint","url":"https://arxiv.org/abs/2603.09002","type":"preprint"},{"label":"Li et al., Security Considerations for Artificial Intelligence Agents, 2026, preprint","url":"https://arxiv.org/abs/2603.12230","type":"preprint"}]},{"category":"evidence","title":"Self-driving laboratory capability and safety review added","change":"Autonomous AI Agents adds Boiko et al.'s demonstration that an LLM-based system (Coscientist) can autonomously plan and execute chemical synthesis, including operating lab hardware, alongside a 2025 Royal Society Open Science review of the safety and security implications of self-driving laboratories.","why":"Documents both a demonstrated autonomous-lab capability and an independent review of the safety questions it raises.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Boiko et al., Nature, 2023","url":"https://doi.org/10.1038/s41586-023-06792-0","type":"paper"},{"label":"Tobias and Wahab, 2025","url":"https://pubmed.ncbi.nlm.nih.gov/40852582/","type":"paper"}]},{"category":"evidence","title":"Skill supply chain vulnerabilities and cascading agent failures documented","change":"Autonomous AI Agents adds a demonstrated backdoored-skill attack that reached the top of a major agent skill repository's popularity charts, describes how corrupted outputs propagate through shared agent memory as cascading failures (OWASP ASI08), and notes CNBC's unverified report that an agent-exclusive social network reached 1.5 million registered agents within days of its January 2026 launch.","why":"Shows that a single compromised skill can propagate through an agent network without further attacker action, and flags an unverified but illustrative growth claim about agent-to-agent platforms.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"The Stack, January 2026","url":"https://www.thestack.technology/a-backdoor-was-the-most-downloaded-skill-for-viral-clawdbot-moltbot-and-why-that-matters/","type":"news"},{"label":"OWASP, Top 10 for Agentic Applications, 2025","url":"https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/","type":"report"},{"label":"CNBC, February 2026","url":"https://www.cnbc.com/2026/02/02/openclaw-open-source-ai-agent-rise-controversy-clawdbot-moltbot-moltbook.html","type":"news"}]},{"category":"evidence","title":"Sycophancy risk in dual-use settings added","change":"LLMs and Information Hazards adds a Sycophancy Problem section: an 11-model Science study found AI systems affirmed users' actions 49% more often than humans, including in scenarios involving deception or illegality, and modeling work suggests repeated validation can drive delusional spiraling even in non-adversarial settings.","why":"Identifies unwarranted affirmation, not just factual inaccuracy, as a failure mode that can reinforce malicious framing and reduce pushback on harmful intent.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Cheng et al., Science, 2026","url":"https://www.science.org/doi/10.1126/science.aec8352","type":"paper"},{"label":"Chandra et al., 2026, preprint","url":"https://arxiv.org/abs/2602.19141","type":"preprint"}]}]},
{"date":"2026-04-18","headline":"OpenAI uplift study detailed; OPCW confirms a chemical evaluation gap","entries":[{"category":"revision","title":"OpenAI's 2024 uplift study expanded with stage-specific detail","change":"LLMs and Information Hazards adds full detail on OpenAI's GPT-4 uplift study (50 biology experts, 50 students, five threat-creation stages), including per-cohort accuracy and completeness scores, and the finding that GPT-4 assistance raised student performance to the expert baseline specifically for the Magnification and Formulation stages.","why":"Gives readers the study's methodology and stage-level results behind a previously summarized uplift finding, including a risk concentration that aggregate results alone would miss.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[]},{"category":"policy","title":"OPCW confirms a chemical-specific AI evaluation gap","change":"LLMs and Information Hazards adds a Chemical Threats section and a CWC Schedule 1/2/3 table, noting that OpenAI's Preparedness Framework v2 uses biological evaluations as a proxy for combined bio/chemical risk, a gap OPCW's AI working group confirmed in March 2026. How AI Amplifies Biosecurity Risk adds the Framework's High and Critical thresholds and 2021 baseline, cross-referencing the section.","why":"An independent international body confirmed that chemical-specific risks are not separately evaluated under the industry's leading capability framework.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"OPCW Scientific Advisory Board, Temporary Working Group on AI, 2026","url":"https://www.opcw.org/media-centre/news/2026/03/opcw-releases-landmark-report-ai-and-chemical-weapons-convention","type":"policy"},{"label":"OPCW, Annex on Chemicals","url":"https://www.opcw.org/chemical-weapons-convention/annexes/annex-chemicals/annex-chemicals","type":"policy"}]},{"category":"developments","title":"OpenAI's Model Spec sets CBRN safety as a root-level rule","change":"LLMs and Information Hazards adds OpenAI's Model Spec (December 2025), which makes CBRN safety a hardcoded, root-level commitment that cannot be overridden by system messages, developers, or users, distinguishing it from instructable behaviors operators can adjust within bounds.","why":"Documents a developer's policy architecture that prevents API customers from overriding CBRN safety by claiming a legitimate use case.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"OpenAI, Model Spec, 2025","url":"https://model-spec.openai.com/2025-12-18.html","type":"announcement"}]}]},
{"date":"2026-04-20","headline":"New chapter: The Chemical Weapons Convention and OPCW","entries":[{"category":"content","title":"New chapter: The Chemical Weapons Convention and OPCW","change":"Adds a chapter on the Chemical Weapons Convention: treaty structure, the OPCW's verification regime of routine and challenge inspections, the Schedule classification system, the destruction of all 72,304 metric tonnes of declared stockpiles, and enforcement failures in Syria and the Novichok poisonings in Salisbury and against Navalny. The comparison with the BWC distinguishes the treaties' verification regimes.","why":"Gives the handbook a full treatment of the one arms-control treaty with a working verification regime, as a comparison case for the unverified Biological Weapons Convention.","chapters":[["The Chemical Weapons Convention and OPCW","/governance/chemical-weapons-convention.html"],["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"],["Executive Summary","/executive-summary.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"]],"sources":[{"label":"OPCW, Chemical Weapons Convention overview","url":"https://www.opcw.org/chemical-weapons-convention","type":"policy"},{"label":"OPCW, chemical weapons stockpile destruction, 2023","url":"https://www.opcw.org/our-work/eliminating-chemical-weapons","type":"policy"},{"label":"Arms Control Association, 2021","url":"https://www.armscontrol.org/act/2021-09/features/syria-russia-and-global-chemical-weapons-crisis","type":"report"},{"label":"RUSI, Novichok attack analysis","url":"https://www.rusi.org/explore-our-research/publications/commentary/syria-and-salisbury-continuing-battle-maintain-trust-evidence-illegal-use-chemical-weapons","type":"report"}]}]},
{"date":"2026-04-22","headline":"Biosecurity Data Levels and dataset-level controls","entries":[{"category":"revision","title":"Biosecurity Data Levels and dataset-level controls","change":"How AI Amplifies Biosecurity Risk holds that for models trained on pathogen sequence, host range, virulence, or immune-evasion data, the dataset is often the most actionable control point, and the Biosecurity Data Levels proposal sets out a tiered framework for it.","why":"Explains tiered controls for infectious-disease datasets.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[]}]},
{"date":"2026-04-23","headline":"AI capability framing updated: uplift findings, GPT-4, and AlphaFold figures; SecureBio Detection: from concept to early operational biosurveillance; Dual-use H5N1 case study and biosecurity fellowship link updated; Biosafety containment tiers and GISRS network history expanded; OPCW working group report: AI-enabled chemical weapons risks","entries":[{"category":"revision","title":"AI capability framing updated: uplift findings, GPT-4, and AlphaFold figures","change":"How AI Amplifies Biosecurity Risk notes the 2024 mild-uplift findings are superseded by OpenAI's April 2025 Preparedness Framework v2, which expects upcoming models to reach High biological capability. AI and Machine Learning Fundamentals qualifies GPT-4's parameter count as unconfirmed and raises AlphaFold's protein count to over 214 million; Digital Biosurveillance now lists pathogen-agnostic metagenomics under SecureBio Detection.","why":"Keeps the chapter's capability assessment aligned with developer evaluations and public estimates that have moved past the 2024 baseline.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["AI and Machine Learning Fundamentals","/ai-biosecurity/index.html"],["Digital Biosurveillance: From Wearables to Real-Time Genomics","/ai-biosecurity/digital-biosurveillance.html"]],"sources":[]},{"category":"developments","title":"SecureBio Detection: from concept to early operational biosurveillance","change":"AI for Biosecurity Defense renames the Nucleic Acid Observatory to SecureBio Detection, now processing over 270 billion read pairs, and describes the metagenomic system as moving from concept to early operational status.","why":"Reflects the program's rebrand and its move from concept to early operational status.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"SecureBio, Detection biosurveillance program","url":"https://securebio.org/detection","type":"announcement"}]},{"category":"revision","title":"Dual-use H5N1 case study and biosecurity fellowship link updated","change":"Dual-Use Research of Concern recounts the 2011 creation of airborne-transmissible H5N1 variants by two research groups and the NSABB review that followed, with the virus's approximately 50% human case fatality rate. Building a Biosecurity Career lists the Emerging Leaders in Biosecurity fellowship.","why":"Frames the 2011 H5N1 controversy through the DURC oversight lens and keeps the fellowship program's reference link current.","chapters":[["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Building a Biosecurity Career: Federal Agencies, Fellowships, and AI-Bio Roles","/governance/biosecurity-careers.html"]],"sources":[]},{"category":"revision","title":"Biosafety containment tiers and GISRS network history expanded","change":"Laboratory Biosafety and Biosecurity details primary and secondary containment barriers and BSL-4 requirements, citing NIH and CDC's Biosafety in Microbiological and Biomedical Laboratories manual. Outbreak Detection and Surveillance adds that WHO established the Global Influenza Surveillance and Response System in 1952, now comprising about 160 National Influenza Centres in over 130 countries.","why":"Grounds containment-tier descriptions and the influenza surveillance network's scale in their primary reference sources.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[]},{"category":"policy","title":"OPCW working group report: AI-enabled chemical weapons risks","change":"The Chemical Weapons Convention and OPCW adds AI-Enabled Chemical Design Risk and AI-Enhanced Chemical Detection sections, citing the OPCW Scientific Advisory Board's Temporary Working Group report (SAB/REP/1/26, March 2026) on retrosynthesis planning, toxicity prediction, and misuse indicators. How AI Amplifies Biosecurity Risk adds that AI toxicity models reached balanced accuracies of 80 to 95% for selected acute-toxicity endpoints.","why":"The OPCW's scientific advisory body released an authoritative assessment of AI-enabled risks at the chemical-biological interface, addressing gaps that biological-only evaluation frameworks leave unaddressed.","chapters":[["The Chemical Weapons Convention and OPCW","/governance/chemical-weapons-convention.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Luechtefeld et al., Toxicological Sciences, 2018","url":"https://doi.org/10.1093/toxsci/kfy152","type":"paper"},{"label":"OPCW Scientific Advisory Board, SAB/REP/1/26, March 2026","url":"https://www.opcw.org/sites/default/files/documents/2026/03/Final%20Report%20of%20the%20SAB%27s%20TWG%20on%20AI%20FINAL%20VERSION.pdf","type":"policy"}]}]},
{"date":"2026-05-03","headline":"OSTP biosecurity framework revision deadline lapses; legislative tracks noted; FY2026 NDAA data-governance provision for DoD biological data added; Structure-prediction tools fail to catch a validated SARS-CoV-2 binding mutant","entries":[{"category":"policy","title":"OSTP biosecurity framework revision deadline lapses; legislative tracks noted","change":"How AI Amplifies Biosecurity Risk reports that the 90-day deadline for OSTP to revise the paused 2024 framework under Executive Order 14292 passed in August 2025 without a published replacement, and adds that Congress has introduced H.R. 3029 (NIST screening standards) and the Biosecurity Modernization and Innovation Act (S. 3741, mandatory screening through Commerce).","why":"Tracks the status of federal biosecurity data governance after the prior framework's revision deadline lapsed without a replacement.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Executive Order 14292","url":"https://www.federalregister.gov/documents/2025/05/08/2025-08266/improving-the-safety-and-security-of-biological-research","type":"policy"},{"label":"H.R. 3029 bill text","url":"https://www.congress.gov/bill/119th-congress/house-bill/3029/text","type":"policy"}]},{"category":"policy","title":"FY2026 NDAA data-governance provision for DoD biological data added","change":"AI-Enabled Pathogen Design adds the FY2026 NDAA's Section 245, which directs the Department of Defense to set requirements, including tiered cybersecurity safeguards and access controls, for how biological data from DoD-funded research is collected and stored for AI use.","why":"Flags an early federal data-governance requirement for DoD-funded biological data, though it covers DoD research only, not the broader open-science ecosystem the chapter's data-level framework targets.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[]},{"category":"evidence","title":"Structure-prediction tools fail to catch a validated SARS-CoV-2 binding mutant","change":"AI-Enabled Pathogen Design adds a finding that AlphaFold 3, AF3Complex, and SpatialPPIv2 all failed to detect an experimentally validated SARS-CoV-2 binding mutant, even for viruses they were trained on, concluding that no single screening layer is sufficient.","why":"Shows that inference-time protein-protein interaction filters cannot substitute for dedicated biosafety screening.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[]}]},
{"date":"2026-05-04","headline":"AlphaGenome added to AI-Enabled Pathogen Design; Anthropic reports training-time backdoor attack on classifiers; OECD synthetic biology and AI assessment added","entries":[{"category":"content","title":"AlphaGenome added to AI-Enabled Pathogen Design","change":"AI-Enabled Pathogen Design adds a Regulatory Genomics section on AlphaGenome (Google DeepMind, 2025), which predicts how noncoding DNA variants affect gene regulation, RNA splicing, and transcription factor binding across tissue types from up to one million base pairs. It states that engineering noncoding mutations to enhance pathogen transmissibility remains theoretical and undemonstrated.","why":"Extends the chapter's coverage from protein structure prediction to the regulatory genome, with the same dual-use caveat applied to current tools.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[{"label":"Google DeepMind, 2025","url":"https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/","type":"announcement"}]},{"category":"developments","title":"Anthropic reports training-time backdoor attack on classifiers","change":"Red-Teaming AI Systems adds that Anthropic research found inserting about 32 poisoned examples into classifier training data installs a backdoor with a capability decrease small enough to pass standard deployment testing, a training-time attack that runtime red-teaming does not detect (Bowers et al., 2026).","why":"Identifies a training-time attack surface that runtime jailbreak testing alone would miss, with implications for how classifier defenses should be evaluated.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Bowers et al., Anthropic alignment science blog, 2026","url":"https://alignment.anthropic.com/2026/backdooring-classifiers/","type":"announcement"}]},{"category":"policy","title":"OECD synthetic biology and AI assessment added","change":"Policy Frameworks for AI-Bio Convergence adds the OECD's Synthetic Biology, AI and Automation Assessment (December 2025), a forward-looking technology assessment covering data supply chains, human oversight requirements, and biosafety and biosecurity governance across OECD member states.","why":"Gives the chapter an intergovernmental reference point for AI and synthetic biology convergence governance.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"OECD, Synthetic Biology, AI and Automation Assessment, 2025","url":"https://www.oecd.org/en/publications/synthetic-biology-ai-and-automation_12158721-en.html","type":"policy"}]}]},
{"date":"2026-05-07","headline":"International AI Safety Report 2026: biological risk findings; Bobier et al. 2025: DURC policy gaps for AI-driven chemical design; CAISI pre-deployment testing expands to Google DeepMind, Microsoft, xAI; Short-fragment DNA screening gap clarified","entries":[{"category":"policy","title":"International AI Safety Report 2026: biological risk findings","change":"How AI Amplifies Biosecurity Risk adds a subsection on the International AI Safety Report 2026 (Bengio et al., February 2026), citing its VCT finding that o3 outperformed 94% of virology experts. The report's survey of 375 biological AI tools found 23% of top performers carry high misuse potential, 61.5% of those fully open source, and only 3% have safeguards.","why":"The most authoritative multilateral assessment of AI biosecurity risk to date finds capability has crossed a threshold that open-source safeguards have not kept pace with.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Bengio et al., International AI Safety Report 2026","url":"https://www.gov.uk/government/publications/international-ai-safety-report-2026","type":"policy"}]},{"category":"evidence","title":"Bobier et al. 2025: DURC policy gaps for AI-driven chemical design","change":"How AI Amplifies Biosecurity Risk and Dual-Use Research of Concern cite Bobier et al., 2025 (Journal of Medical Ethics), arguing that current DURC policy, built around the seven Fink Report categories, does not account for AI-driven pharmaceutical and chemical design such as the MegaSyn case, and calling for broadened DURC scope.","why":"Applies dual-use research oversight, designed for pathogens, to AI-accelerated pharmaceutical and chemical design work it was not built to cover.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"sources":[{"label":"Bobier et al., Journal of Medical Ethics, 2025","url":"https://doi.org/10.1136/jme-2025-110750","type":"paper"}]},{"category":"policy","title":"CAISI pre-deployment testing expands to Google DeepMind, Microsoft, xAI","change":"How AI Amplifies Biosecurity Risk reports that the U.S. Center for AI Standards and Innovation's pre-deployment testing program expanded in May 2026 when Google DeepMind, Microsoft, and xAI joined OpenAI and Anthropic in providing frontier models, including versions with reduced safeguards, for government evaluation. CAISI reports completing more than 40 such evaluations through the interagency TRAINS Taskforce.","why":"Tracks which frontier AI developers now submit models, including reduced-safeguard versions, for government biosecurity evaluation.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"NIST, CAISI frontier AI national security testing agreements, May 2026","url":"https://www.nist.gov/news-events/news/2026/05/caisi-signs-agreements-regarding-frontier-ai-national-security-testing","type":"policy"}]},{"category":"content","title":"Short-fragment DNA screening gap clarified","change":"DNA Synthesis Screening adds an entry on whether orders under 200 base pairs are screened: not reliably, since the IGSC baseline covers 200+ bp orders while oligonucleotides can be assembled into larger constructs. It notes the 2024 OSTP framework's planned move to 50-nucleotide window screening by October 13, 2026 is now under revision after Executive Order 14292.","why":"Gives readers a direct answer on a screening gap that oligonucleotide assembly could otherwise exploit.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[]}]},
{"date":"2026-05-09","headline":"GPT-5 autonomously ran cell-free protein synthesis experiments at Ginkgo Bioworks","entries":[{"category":"evidence","title":"GPT-5 autonomously ran cell-free protein synthesis experiments at Ginkgo Bioworks","change":"Cloud Labs and Automated Biology and Autonomous AI Agents report that OpenAI's GPT-5 proposed and analyzed cloning experiments in December 2025 with human execution, achieving a 79-fold improvement in molecular cloning efficiency. By February 2026, GPT-5 connected directly to Ginkgo Bioworks' cloud laboratory and autonomously ran over 36,000 cell-free protein synthesis experiments across six rounds without human execution.","why":"Marks a shift from AI-assisted lab work under human execution to fully autonomous experimental cycles in a commercial cloud laboratory.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"],["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"OpenAI, GPT-5 lowers protein synthesis cost, 2026","url":"https://openai.com/index/gpt-5-lowers-protein-synthesis-cost/","type":"announcement"},{"label":"OpenAI and Ginkgo Bioworks, 2026, preprint","url":"https://www.biorxiv.org/content/10.64898/2026.02.05.703998v1","type":"preprint"}]}]},
{"date":"2026-05-15","headline":"GovAI: coding agents lower the skill barrier for BAIM fine-tuning","entries":[{"category":"evidence","title":"GovAI: coding agents lower the skill barrier for BAIM fine-tuning","change":"How AI Amplifies Biosecurity Risk adds a third risk channel beyond the LLM/BDT binary: a March 2026 GovAI experiment found a non-biology expert used Claude Code to replicate expert-level fine-tuning of Evo 2 in one weekend for about $760, encountering no refusals. Training-set perplexity improved significantly, but the held-out test-set improvement did not reach statistical significance.","why":"Shows fine-tuning safeguards premised on technical difficulty can be eroded by coding agents, prompting a call to evaluate models for BAIM modification risk, not just use.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Righetti, Lukosiute, and Black, GovAI, April 2026","url":"https://www.governance.ai/analysis/coding-agents-are-changing-the-biosecurity-risk-landscape","type":"report"}]}]},
{"date":"2026-05-21","headline":"Multi-agent AI (Robin, Co-Scientist) extends into open-ended biology; PDF edition added","entries":[{"category":"evidence","title":"Robin and Co-Scientist: multi-agent AI extends into open-ended biology","change":"Autonomous AI Agents adds two May 2026 Nature papers, FutureHouse's Robin (Ghareeb et al.) and Google's Co-Scientist (Gottweis et al.), reporting multi-agent systems that generated hypotheses and analyzed experiments in open-ended mammalian biology, including a candidate drug (ripasudil) Robin proposed for macular degeneration. It notes the assumption that autonomous agents operate within predefined domains no longer describes the published frontier.","why":"Extends the chapter's constrained-domain framing for autonomous lab agents now that published systems operate across both chemistry and mammalian biology.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Ghareeb et al., Nature, 2026, accepted manuscript","url":"https://doi.org/10.1038/s41586-026-10652-y","type":"paper"},{"label":"Gottweis et al., Nature, 2026, accepted manuscript","url":"https://doi.org/10.1038/s41586-026-10644-y","type":"paper"}]},{"category":"evidence","title":"Prospective respiratory early-warning system tested across 50 states","change":"AI for Biosecurity Defense adds a transfer-learning early warning system, tested prospectively across all 50 U.S. states in the 2024 to 2025 respiratory season, combining Google search trends with epidemiological data. It detected 98.0% of outbreak onsets and 97.0% of peaks, with positive predictive values above 82% (Garrido-Garcia et al., 2026).","why":"Shows early-warning systems become more credible for biosecurity use when digital signals are fused with epidemiological data and tested prospectively rather than retrospectively.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"Garrido-Garcia et al., 2026","url":"https://doi.org/10.1038/s41467-026-72655-7","type":"paper"}]},{"category":"evidence","title":"DeepAVC framework identifies broad-spectrum antiviral candidate","change":"AI for Biosecurity Defense adds Kang et al., 2026, describing DeepAVC, an LLM-powered antiviral discovery framework combining phenotype-based and target-based prediction, which identified MNS as a broad-spectrum antiviral candidate with greater efficacy than Sisunatovir in the authors' in vitro and in vivo tests. The chapter notes this is early-stage defensive research, not stockpiled countermeasure readiness.","why":"Documents AI-assisted antiviral discovery as defensive evidence while flagging that dual-use design capability carries the same caveat.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"Kang et al., 2026","url":"https://doi.org/10.1038/s42003-026-10291-z","type":"paper"}]},{"category":"milestone","title":"Downloadable PDF edition added","change":"The handbook adds a downloadable PDF edition alongside the web chapters.","why":"Gives readers an offline, printable version of the full handbook.","chapters":[["References","/references.html"]],"sources":[]}]},
{"date":"2026-05-25","headline":"LLMs evaluated for recognizing Category A bioterrorism syndromes","entries":[{"category":"evidence","title":"LLMs evaluated for recognizing Category A bioterrorism syndromes","change":"AI for Biosecurity Defense adds a study evaluating frontier LLMs on recognizing CDC Category A bioterrorism agent syndromes under idealized vignette conditions, suggesting a potential role as diagnostic support and training tools with appropriate governance and validation.","why":"Gives the chapter primary evidence on LLM performance in bioterrorism syndrome recognition.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"Armitage, 2026","url":"https://doi.org/10.1080/23779497.2026.2643956","type":"paper"}]}]},
{"date":"2026-05-26","headline":"NIST safe-proxy evaluation of AI-assisted protein design risk; NIST inter-tool benchmark for DNA synthesis screening; Built-in AI-bio safeguards remain immature, correspondence finds; Federated infectious disease surveillance reviewed in Nature Medicine","entries":[{"category":"evidence","title":"NIST safe-proxy evaluation of AI-assisted protein design risk","change":"AI-Enabled Pathogen Design and Red-Teaming AI Systems add a NIST safe-proxy testing, evaluation, validation, and verification study, which found that AI-assisted protein design tools could generate synthetic homologs with similar predicted structures but did not reliably preserve biological activity while evading screening.","why":"Gives both chapters primary evidence on the current reliability limits of AI-assisted protein-design risk to inform screening policy.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Ikonomova et al., 2025","url":"https://www.nist.gov/publications/experimental-evaluation-ai-driven-protein-design-risks-using-safe-biological-proxies","type":"report"}]},{"category":"evidence","title":"NIST inter-tool benchmark for DNA synthesis screening","change":"DNA Synthesis Screening adds a NIST blinded 999-fragment benchmark comparing six sequence-screening tools, reporting baseline performance above 95% sensitivity and 97% accuracy, with disagreements traced mainly to differing sequence-of-concern definitions or algorithmic methods.","why":"Gives the chapter a primary benchmark baseline for the sequence-screening layer, supporting harmonized datasets as governance infrastructure.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Laird et al., 2025","url":"https://doi.org/10.1177/15356760251401228","type":"paper"}]},{"category":"evidence","title":"Built-in AI-bio safeguards remain immature, correspondence finds","change":"Adds a Nature Biotechnology correspondence finding that watermarking, alignment, anti-jailbreak methods, and unlearning require domain-specific research before they qualify as dependable biosecurity controls.","why":"Gives the chapter primary evidence on the current maturity of built-in AI-bio safeguards and the latest tracked policy trends.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Wang et al., 2025","url":"https://doi.org/10.1038/s41587-025-02650-8","type":"paper"},{"label":"NTI, AIxBio Horizon Scan, Spring 2026","url":"https://www.nti.org/analysis/articles/aixbio-horizon-scan-spring-2026/","type":"report"}]},{"category":"evidence","title":"Federated infectious disease surveillance reviewed in Nature Medicine","change":"Digital Biosurveillance adds a review describing federated infectious disease surveillance, which keeps data local while running analysis across distributed datasets when outbreak data cannot be pooled across borders. The review notes adoption remains limited, requiring shared protocols, transparent governance, and local analytic capacity.","why":"Gives the chapter a primary source for federated surveillance as one architectural option for data-sharing barriers.","chapters":[["Digital Biosurveillance: From Wearables to Real-Time Genomics","/ai-biosecurity/digital-biosurveillance.html"]],"sources":[{"label":"Khurana et al., 2026","url":"https://doi.org/10.1038/s41591-026-04351-4","type":"paper"}]}]},
{"date":"2026-06-23","headline":"Los Alamos biosurveillance gateway and Sandia biorisk models cited","entries":[{"category":"developments","title":"Los Alamos and Sandia biosurveillance and biorisk tools cited","change":"Digital Biosurveillance adds Los Alamos National Laboratory's Biosurveillance Gateway as an institutional example combining environmental surveillance, genomics, multi-omics, predictive modeling, and decision support for biothreat detection. Laboratory Biosafety and Biosecurity and Further Reading add Sandia's BioRisk Assessment Models for prioritizing site-specific biosafety and biosecurity risks.","why":"Gives readers concrete institutional examples of integrated detection and risk-prioritization tools.","chapters":[["Digital Biosurveillance: From Wearables to Real-Time Genomics","/ai-biosecurity/digital-biosurveillance.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"],["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"Los Alamos National Laboratory, Biosurveillance Gateway","url":"https://www.lanl.gov/engage/collaboration/biosurveillance-gateway","type":"other"},{"label":"Sandia National Laboratories, GCBS BioRisk Assessment Models","url":"https://gcbs.sandia.gov/tools/biosecurity-biosafety-ram/","type":"other"}]}]},
{"date":"2026-06-26","headline":"PPE, respirator, and Far-UVC environmental-control guidance expanded","entries":[{"category":"revision","title":"PPE, hierarchy of controls, and Far-UVC environmental controls","change":"Laboratory Biosafety and Biosecurity and Medical Countermeasures and Biodefense add CDC/NIOSH and OSHA sourcing on the hierarchy of controls, PPE selection, and respirator programs, plus germicidal UV and Far-UVC as supplemental air treatment, not replacements for ventilation or PPE. New citations include a null-result randomized GUV trial in long-term care and a 2023 HHS OIG finding that the Strategic National Stockpile could not meet 2020 PPE demand.","why":"Grounds protective-equipment guidance in CDC/NIOSH, OSHA, and peer-reviewed sources rather than general practice.","chapters":[["Glossary of Biosecurity Terms","/appendices/glossary.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"],["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"CDC/NIOSH, Hierarchy of Controls","url":"https://www.cdc.gov/niosh/hierarchy-of-controls/about/index.html","type":"policy"},{"label":"OSHA, 29 CFR 1910.134","url":"https://www.osha.gov/laws-regs/regulations/standardnumber/1910/1910.134","type":"policy"},{"label":"CDC/NIOSH, Germicidal Ultraviolet Guidance","url":"https://www.cdc.gov/niosh/ventilation/germicidal-ultraviolet/index.html","type":"policy"},{"label":"Shoubridge et al., 2025","url":"https://doi.org/10.1001/jamainternmed.2025.3388","type":"paper"},{"label":"HHS OIG, November 2023","url":"https://www.oversight.gov/sites/default/files/documents/reports/2023-11/42002028.pdf","type":"report"},{"label":"Verbeek et al., 2020","url":"https://doi.org/10.1002/14651858.CD011621.pub4","type":"paper"}]}]},
{"date":"2026-06-29","headline":"Real-room translation limits added for Far-UVC delivery","entries":[{"category":"evidence","title":"Real-room translation limits added for Far-UVC delivery","change":"Medical Countermeasures and Biodefense adds a source on real-room limits to Far-UVC delivery, noting that partial room irradiation and factors like ventilation and air mixing affect whether chamber-demonstrated pathogen inactivation translates to occupied spaces.","why":"Reinforces that chamber efficacy for Far-UVC does not by itself establish real-world disease-transmission reduction.","chapters":[["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[{"label":"Memarzadeh et al., 2010","url":"https://doi.org/10.1016/j.ajic.2010.04.208","type":"paper"}]}]},
{"date":"2026-07-17","headline":"Assurance framework for AI-enabled AMR analytics","entries":[{"category":"evidence","title":"Assurance framework for AI-enabled AMR analytics","change":"AI for Biosecurity Defense adds a six-question assurance framework for AI tools that predict antimicrobial resistance, covering reference-database dependence, the gap between a resistance-gene call and a confirmed resistant infection, and geographic or temporal generalization. Halicin coverage cites the original Stokes et al. (2020) study.","why":"Both cited resistance-prediction tools are described by their own authors as research prototypes, not validated diagnostics; the framework gives readers questions to ask before treating either as clinically ready.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"Nguyen et al., 2026","url":"https://doi.org/10.1038/s41467-026-69934-8","type":"paper"},{"label":"Mollerus et al., 2026","url":"https://doi.org/10.1038/s44259-026-00219-2","type":"paper"},{"label":"Doyle et al., 2020","url":"https://doi.org/10.1099/mgen.0.000335","type":"paper"},{"label":"Hu et al., 2024","url":"https://doi.org/10.1093/bib/bbae206","type":"paper"},{"label":"Stokes et al., 2020","url":"https://doi.org/10.1016/j.cell.2020.01.021","type":"paper"}]}]},
{"date":"2026-07-19","headline":"A testable, consortium-wide definition of 'sequence of concern'","entries":[{"category":"evidence","title":"A testable definition of 'sequence of concern'","change":"DNA Synthesis Screening adds the Sequence Biosecurity Risk Consortium's Biosecurity Flag Rubric, paired with curated test sets that four independently developed screening tools evaluated. Applying the rubric to roughly 1.1 million sequences reduced disputed classifications by 44.3% for controlled viruses and 10.7% overall, a measure of consortium agreement rather than real-world risk reduction.","why":"Organism control lists alone do not resolve which individual sequences should be flagged; the consortium's rubric and test sets give screening tool developers a shared, testable standard to converge toward.","chapters":[["Glossary of Biosecurity Terms","/appendices/glossary.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Alexanian et al., 2026","url":"https://doi.org/10.3389/fbioe.2026.1830842","type":"paper"}]}]},
{"date":"2026-07-28","headline":"Statistical power limits on null LLM-uplift findings; Biological design tool scope, from molecules to whole-body simulators; UK AISI and SecureBio evaluations of Constitutional Classifiers","entries":[{"category":"evidence","title":"Statistical power limits on null LLM-uplift findings","change":"LLMs and Information Hazards adds a caveat that RAND, OpenAI, and Hong et al.'s null results on biological uplift may reflect underpowered samples rather than a true absence of effect. In Hong et al.'s trial, LLM-assisted cell-culture success was 68.8% versus 55.3% without, a gap that did not reach statistical significance at n=153.","why":"A 2024 statistical framework for evaluating AI systems notes that failing to reach significance in a modest sample does not prove no meaningful effect exists, which bears on capability-threshold governance decisions.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Miller, 2024, preprint","url":"https://arxiv.org/abs/2411.00640","type":"preprint"}]},{"category":"revision","title":"Biological design tool scope, from molecules to whole-body simulators","change":"AI and Machine Learning Fundamentals and AI-Enabled Pathogen Design broaden the biological design tool category beyond molecule-level tools (AlphaFold, RFdiffusion, ESM-3) to include multi-scale simulators that model biology from molecular data through organ-system or whole-body response.","why":"Existing worked examples of biological design tools operated only at the molecule or protein level, leaving organism-level physiological simulators outside the same governance framework.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["AI and Machine Learning Fundamentals","/ai-biosecurity/index.html"]],"sources":[]},{"category":"developments","title":"UK AISI and SecureBio evaluations of Constitutional Classifiers","change":"LLMs and Information Hazards and Red-Teaming AI Systems add UK AI Security Institute's Boundary Point Jailbreaking method, which produced universal jailbreaks against Constitutional Classifiers, and SecureBio's external review, which found high refusal coverage does not establish adversarial robustness against highly skilled attackers.","why":"Anthropic's own red-teaming had reported no universal jailbreak; independent evaluators subsequently found a working bypass method and concluded that jailbreak-resistance claims are conditional on attacker skill and time.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"]],"sources":[{"label":"UK AI Security Institute, February 2026","url":"https://www.aisi.gov.uk/blog/boundary-point-jailbreaking-a-new-way-to-break-the-strongest-ai-defences","type":"policy"},{"label":"SecureBio, July 2026","url":"https://securebio.org/resources/anthropic_feb_2026_risk_review.pdf","type":"report"}]}]},
{"date":"2026-07-30","headline":"Pathogen genomics training resource; A testable biosecurity project portfolio","entries":[{"category":"content","title":"Pathogen genomics training resource","change":"Further Reading adds a free, self-paced course series on pathogen genomics for public health, covering conceptual foundations, genomic epidemiology, and sequencing strategies for surveillance and outbreak response.","why":"Gives readers a structured, no-cost training path for the genomic-surveillance skills the handbook discusses.","chapters":[["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"Pathogen Genomics for Public Health","url":"https://learn.hms.harvard.edu/pathogen-genomics-public-health","type":"other"}]},{"category":"content","title":"A testable biosecurity project portfolio","change":"The Future of Biosecurity adds an evidence-gated portfolio for a proposed ten-project biosecurity agenda (Bakerlee, 2026), assigning each project an evidence class and a minimum gate before it can inform policy. Synthetic Biology and Democratization and the Glossary add mirror-life risk coverage, citing a 2024 Science assessment and a 2026 UN Scientific Advisory Board brief on research red lines.","why":"The agenda's author frames the ten projects as hypotheses, not settled priorities; the portfolio's evidence classes and gates keep proposals from becoming policy or emergency guidance before they are tested.","chapters":[["Glossary of Biosecurity Terms","/appendices/glossary.html"],["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["Executive Summary","/executive-summary.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[{"label":"Bakerlee, 2026","url":"https://defensesindepth.bio/10-big-projects-for-reducing-bio-x-risk/","type":"other"},{"label":"Adamala et al., 2024","url":"https://doi.org/10.1126/science.ads9158","type":"paper"},{"label":"UN Scientific Advisory Board, 2026","url":"https://www.un.org/scientific-advisory-board/en/mirror-life","type":"policy"},{"label":"Wittmann et al., 2025","url":"https://doi.org/10.1126/science.adu8578","type":"paper"},{"label":"Richards, 2025, preprint","url":"https://arxiv.org/abs/2510.17828","type":"preprint"},{"label":"NIOSH, 2024","url":"https://www.cdc.gov/niosh/bulletin/2024/tips-for-managing-personal-protective-equipment-in-your-stockpile-respirator-selection-and-purchase.html","type":"policy"}]},{"category":"revision","title":"Five analytical layers for AI biosecurity red-teaming evaluation","change":"Red-Teaming AI Systems covers five analytical layers (capability measurement, human and agent uplift, safeguard effectiveness, operational validation, and governance), and adds OpenAI's and Anthropic's July 2026 system cards, NIST's benchmark-validity reports, and the EU AI Act's Article 55 evaluation requirement.","why":"Frames biosecurity red-teaming as a full assurance process, not a single jailbreak test, so findings on one safeguard version are not assumed to transfer to later models or deployments.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Executive Summary","/executive-summary.html"]],"sources":[{"label":"OpenAI, July 2026","url":"https://deploymentsafety.openai.com/gpt-5-6","type":"announcement"},{"label":"Anthropic, July 2026","url":"https://www.anthropic.com/claude-opus-5-system-card","type":"announcement"},{"label":"Mouton et al., 2024","url":"https://doi.org/10.7249/RRA2977-2","type":"paper"},{"label":"Regulation (EU) 2024/1689, Article 55","url":"https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32024R1689","type":"policy"},{"label":"UK AISI, 2025","url":"https://www.aisi.gov.uk/research/principles-for-evaluating-misuse-safeguards-of-frontier-ai-systems","type":"policy"},{"label":"Sudarshan and Righetti, 2026, technical report","url":"https://www.governance.ai/research-paper/technical-report-towards-a-common-standard-for-evaluating-frontier-ai-safeguards-against-biological-misuse","type":"report"}]},{"category":"evidence","title":"RFdiffusion3 and digital-biology uplift benchmarks","change":"AI-Enabled Pathogen Design adds RFdiffusion3, a developer-reported all-atom protein design model at roughly one-tenth the prior computational cost, validated experimentally only on select DNA-binding and enzyme tasks (Butcher et al., 2025, preprint). How AI Amplifies Biosecurity Risk adds a 2026 preprint reporting higher odds of correct answers with LLM access across eight digital-biology benchmarks, with no physical laboratory endpoint tested.","why":"A developer-authored preprint lowers the cost of designing non-protein binding interactions, and a separate preprint reports substantial novice uplift on digital biology tasks with LLM access.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Quick Reference: All Chapter Summaries (TL;DRs)","/appendices/tldr-compilation.html"],["Executive Summary","/executive-summary.html"]],"sources":[{"label":"Butcher et al., 2025, preprint","url":"https://doi.org/10.1101/2025.09.18.676967","type":"preprint"},{"label":"RFdiffusion3 training and inference code (GitHub)","url":"https://github.com/RosettaCommons/foundry/blob/production/models/rfd3/README.md","type":"dataset"},{"label":"Zhang et al., 2026, preprint","url":"https://arxiv.org/abs/2602.23329","type":"preprint"},{"label":"Vieira et al., 2026, preprint","url":"https://doi.org/10.64898/2026.03.08.710389","type":"preprint"}]}]},
{"date":"2026-08-06","headline":"U.S. Government Policy for Stopping High-Risk Life Sciences Research (July 2026); Wittmann et al., 2026 follow-up on synthesis-screening evasion; Anthropic reports Fable 5 biology-safeguard recalibration","entries":[{"category":"policy","title":"U.S. Government Policy for Stopping High-Risk Life Sciences Research (July 2026)","change":"Gain-of-Function Research and Policy Frameworks for AI-Bio Convergence add the July 28, 2026 U.S. Government Policy for Stopping High-Risk Life Sciences Research, issued under EO 14292 Section 4 with NIH notice NOT-OD-26-101, which prohibits federal support for dangerous gain-of-function research and supersedes the P3CO review process; implementation is scheduled for mid-November 2026.","why":"The 2025 gain-of-function funding pause lacked a documented successor until this July 2026 policy replaced it.","chapters":[["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"HHS, U.S. Government Policy for Stopping High-Risk Life Sciences Research, 2026","url":"https://www.hhs.gov/press-room/stopping-high-risk-life-sciences-research.html","type":"policy"},{"label":"NIH, Notice NOT-OD-26-101, 2026","url":"https://grants.nih.gov/grants/guide/notice-files/NOT-OD-26-101.html","type":"policy"}]},{"category":"evidence","title":"Wittmann et al., 2026 follow-up on synthesis-screening evasion","change":"AI-Enabled Pathogen Design and DNA Synthesis Screening add a 2026 follow-up study that tested fragmented synthetic homologs against four patched screening tools and found them effective against the tested design capabilities, though detection declines at lower sequence similarity; neither study produced physical proteins to confirm retained biological function.","why":"A 2026 follow-up tested whether the patches deployed after the 2025 evasion study held against newer synthetic variants.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Wittmann et al., 2026","url":"https://doi.org/10.3389/fbioe.2026.1858951","type":"paper"}]},{"category":"developments","title":"Anthropic reports Fable 5 biology-safeguard recalibration","change":"Red-Teaming AI Systems adds Anthropic's account of revising Fable 5's classifier constitution and training data, which the developer reported cut biology-related fallbacks to a less capable model by approximately 85% while still routing dual-use professional biology and drug-development requests to Opus 5.","why":"Anthropic published an update on its staged biology-query routing, and the chapter notes the announcement did not report paired false-negative or misuse outcomes.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Anthropic, August 2026","url":"https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards","type":"announcement"}]}]},
{"date":"2026-08-09","headline":"Cyberbiosecurity for Connected Laboratories","entries":[{"category":"content","title":"Cyberbiosecurity for Connected Laboratories","change":"Laboratory Biosafety and Biosecurity adds a new section defining a control model for connected instruments, identities, segmentation, genomic-data integrity, and unified incident and near-miss learning.","why":"Connected laboratory instruments and genomic-data pipelines create cyber risks with biological consequences that the handbook had not previously addressed as a unified control model.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"],["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["Digital Biosurveillance: From Wearables to Real-Time Genomics","/ai-biosecurity/digital-biosurveillance.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"NIST, IR 8432","url":"https://doi.org/10.6028/NIST.IR.8432","type":"report"},{"label":"Crawford et al., 2023","url":"https://doi.org/10.3389/fbioe.2023.1240281","type":"paper"},{"label":"Cova et al., 2026","url":"https://doi.org/10.1177/15356760261453108","type":"paper"},{"label":"WHO, Laboratory Biosafety Manual, Fourth Edition","url":"https://www.who.int/publications-detail-redirect/9789240011311","type":"policy"}]},{"category":"evidence","title":"Gu et al., 2026 modeling study on traveler-targeted genomic surveillance","change":"Outbreak Detection and Surveillance adds a new section on a 2026 modeling study finding that adding traveler surveillance at two high-volume hubs reduced simulated mean global identification lag from 14.1 to 11.5 days without reducing non-hub surveillance capacity; the chapter notes these are modeled estimates, not operational validation.","why":"Sequencing capacity is constrained in many settings, and the model tests whether targeted traveler surveillance can complement decentralized sequencing without new resources.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"Gu et al., 2026","url":"https://doi.org/10.1038/s41467-026-70664-0","type":"paper"}]},{"category":"evidence","title":"Feldman et al., 2026 Know Your Scientist governance proposal","change":"Policy Frameworks for AI-Bio Convergence adds a 2026 peer-reviewed Perspective proposing a Know Your Scientist framework in which research institutions vouch for users of hosted biological-design tools while hosts screen outputs and monitor activity over time; the chapter notes the proposal is an unevaluated governance architecture that open-weight or locally run models would bypass.","why":"Hosted biological-design tools raise oversight questions that current frameworks do not address, and the chapter flags which safeguards remain untested.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Feldman et al., 2026","url":"https://doi.org/10.3389/fmicb.2026.1814993","type":"paper"}]}]},
{"date":"2026-08-11","headline":"AI-bio risk: real-world actors and medical countermeasure readiness","entries":[{"category":"revision","title":"AI-bio risk: real-world actors and medical countermeasure readiness","change":"How AI Amplifies Biosecurity Risk uses a scenario framework grounding threat assessment in actor baseline, pathway, and outcome; Medical Countermeasures and Biodefense adds a section on AI-enabled threat preparedness tied to the National Academies consensus study now in progress.","why":"An AI capability finding alone is not a risk estimate; assessment requires attention to actor capability, pathway feasibility, and constraints documented in the historical and policy literature.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[{"label":"National Academy of Medicine, workshop, August 2026","url":"https://nam.edu/our-work/programs/committee-on-emerging-science-technology-and-innovation-in-health-and-medicine/preparing-for-a-future-of-ai-enabled-biology/","type":"other"},{"label":"National Academies, consensus study, 2026","url":"https://www.nationalacademies.org/projects/CHPP-LSB-25-04","type":"report"},{"label":"Vogel, 2013","url":"https://doi.org/10.1177/0096340212470813","type":"paper"},{"label":"Ben Ouagrham-Gormley, 2012","url":"https://doi.org/10.1162/isec_a_00077","type":"paper"},{"label":"Williams et al., RAND, 2026","url":"https://doi.org/10.7249/RRA4490-1","type":"report"},{"label":"Adalja et al., 2026","url":"https://doi.org/10.1093/ofid/ofag108","type":"paper"}]}]},
{"date":"2026-08-12","headline":"King et al., 2026 evidence of AI-generated functional bacteriophages; BWC verification and the July 2026 high-risk research policy; Historical pathogen categories and current select-agent access controls; One Health surveillance and wastewater early-signal evidence; Biological information access and tacit expertise","entries":[{"category":"evidence","title":"King et al., 2026 evidence of AI-generated functional bacteriophages","change":"AI-Enabled Pathogen Design adds King et al.'s report of experimentally functional bacteriophages generated with genome language models in a supervised bacterial-virus research system, noting the result ends the categorical claim that functional virus design is beyond current capability for every system.","why":"The result changes the AI-capability boundary for pathogen design from theoretical to demonstrated, bounded to bacteriophages in a constrained experimental system.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[{"label":"King et al., 2026","url":"https://doi.org/10.1126/science.aec2657","type":"paper"}]},{"category":"revision","title":"BWC verification and the July 2026 high-risk research policy","change":"Executive Summary, International Governance and the BWC, and Dual-Use Research of Concern update Key Findings to reflect the Biological Weapons Convention's lack of standing verification, current DNA-synthesis screening coverage, and the July 28, 2026 U.S. policy that now governs high-risk life sciences research, replacing the DURC framework that never took effect.","why":"Distinguishes the current high-risk research policy from earlier frameworks and places BWC verification gaps in the context of Soviet and Iraqi violations.","chapters":[["Executive Summary","/executive-summary.html"],["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"sources":[{"label":"White House, USG Policy for Stopping High-Risk Life Sciences Research (PDF), July 2026","url":"https://www.whitehouse.gov/wp-content/uploads/2026/07/USG-Policy-for-Stopping-High-Risk-Life-Sciences-Research_July-2026.pdf","type":"policy"},{"label":"International Gene Synthesis Consortium, Harmonized Screening Protocol v3.0","url":"https://genesynthesisconsortium.org/wp-content/uploads/IGSC-Harmonized-Screening-Protocol-v3.0-1.pdf","type":"other"},{"label":"OPCW, 2026","url":"https://www.opcw.org/media-centre/news/2026/05/deployment-opcw-expert-team-syrian-arab-republic-finds-chemical-weapons","type":"policy"},{"label":"UN Security Council 1540 Committee, national reporting statement, August 2025","url":"https://www.un.org/en/sc/1540/documents/statement.chair.sc.6aug25.pdf","type":"policy"}]},{"category":"revision","title":"Historical pathogen categories and current select-agent access controls","change":"Pathogens of Concern describes the CDC Category A, B, and C framework as historical preparedness context, distinct from the current Federal Select Agent Program, and Laboratory Biosafety and Biosecurity separates Security Risk Assessments for select-agent access from enhanced suitability requirements for Tier 1 agents.","why":"The historical CDC bioterrorism categories are commonly conflated with current regulatory access controls, and the chapters now distinguish the two.","chapters":[["Pathogens of Concern","/foundations/pathogens-of-concern.html"],["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"Select Agents Program, Federal Select Agent Program overview","url":"https://www.selectagents.gov/overview/index.htm","type":"policy"},{"label":"FDA, Ebola virus treatment approval","url":"https://www.fda.gov/drugs/news-events-human-drugs/fda-approves-treatment-ebola-virus","type":"policy"},{"label":"CDC, Andes virus hantavirus overview","url":"https://www.cdc.gov/hantavirus/about/andesvirus.html","type":"policy"}]},{"category":"revision","title":"One Health surveillance and wastewater early-signal evidence","change":"Outbreak Detection and Surveillance describes One Health Quadripartite Collaboration, adds that the 2024 International Health Regulations amendments entered into force September 19, 2025, and cites evidence that wastewater surveillance detected SARS-CoV-2 RNA before the first locally reported clinical cases in one studied city.","why":"Reflects the addition of a fourth partner to the One Health governance structure and gives readers the primary evidence for early wastewater detection.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"WHO, consolidated IHR text","url":"https://apps.who.int/gb/bd/pdf_files/IHR_2014-2022-2024-en.pdf","type":"policy"},{"label":"Medema et al., 2020","url":"https://doi.org/10.1021/acs.estlett.0c00357","type":"paper"}]},{"category":"revision","title":"Biological information access and tacit expertise","change":"How AI Amplifies Biosecurity Risk distinguishes written instructions from contributory expertise, noting that detailed model output does not by itself confer laboratory proficiency.","why":"Separates biological information access from hands-on laboratory competence.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[]}]},
{"date":"2026-08-16","headline":"Evidence-interpretation guardrails added across five chapters","entries":[{"category":"revision","title":"Evidence-interpretation guardrails added across five chapters","change":"How AI Amplifies Biosecurity Risk, The Biological Threat Landscape, International Governance and the BWC, Medical Countermeasures and Biodefense, and Outbreak Detection and Surveillance each add a methodological section: mapping uplift measurements to threat-pathway stages, avoiding single-label threat categorization, distinguishing BWC norm from verification, tracing countermeasure readiness stages, and requiring surveillance signals to reach a defined response pathway.","why":"Gives readers explicit criteria for interpreting evidence in each domain instead of single-label conclusions about risk, threat category, treaty compliance, readiness, or detection.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["The Biological Threat Landscape","/foundations/threat-landscape.html"],["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[]}]},
{"date":"2026-08-18","headline":"Post-training safety layers clarified; synthesis-screening chokepoint narrowing","entries":[{"category":"revision","title":"Capability post-training distinguished from safety post-training","change":"How AI Amplifies Biosecurity Risk, AI and Machine Learning Fundamentals, and LLMs and Information Hazards separate capability post-training, safety-behavior post-training, and use-time deployment safeguards into distinct evaluation layers, citing DeepSeek-R1, an ICLR process-supervision study, an ICML preference-labeling study, and the Tulu 3 pipeline as domain-specific evidence that does not establish biosecurity transfer.","why":"Reward hacking and domain-specific reasoning results are often read as evidence about safety; the chapters now require each evaluation to name which layer it actually tested.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["AI and Machine Learning Fundamentals","/ai-biosecurity/index.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Guo et al., 2025","url":"https://doi.org/10.1038/s41586-025-09422-z","type":"paper"},{"label":"Lightman et al., 2024","url":"https://proceedings.iclr.cc/paper_files/paper/2024/hash/aca97732e30bcf1303bc22ac3924fd16-Abstract-Conference.html","type":"paper"},{"label":"Lee et al., 2024","url":"https://proceedings.mlr.press/v235/lee24t.html","type":"paper"},{"label":"Lambert et al., 2025","url":"https://openreview.net/forum?id=i1uGbfHHpH","type":"paper"}]},{"category":"revision","title":"DNA synthesis screening's narrowing chokepoint","change":"DNA Synthesis Screening adds that benchtop synthesizers enabling on-site synthesis and AI-assisted design of sequence-distant candidates both challenge the sequence-and-customer screening chokepoint, alongside the approximately 20% of global synthesis capacity that operates outside voluntary screening frameworks.","why":"Documents how the two-layer screening defense is narrowing as decentralized synthesis technology and design tools advance.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[]}]},
{"date":"2026-08-25","headline":"Nine chapters read in the handbook's formal register","entries":[{"category":"revision","title":"Contractions expanded to formal register across nine chapters","change":"Negative contractions such as can't, doesn't, and isn't are expanded to their full forms across nine chapters, matching the handbook's formal third-person register; quoted dialogue and code are left unchanged.","why":"Keeps prose register consistent for professional readers by removing informal contractions from chapter text.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"],["Gene Drives and Environmental Biosecurity","/biotechnology/gene-drives.html"],["Pathogens of Concern","/foundations/pathogens-of-concern.html"],["What Is Biosecurity?","/foundations/what-is-biosecurity.html"],["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"],["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"],["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[]}]},
{"date":"2026-08-28","headline":"Guerra et al., 2026 RAND report on layered AI-bio defense mitigations","entries":[{"category":"evidence","title":"Guerra et al., 2026 RAND report on layered AI-bio defense mitigations","change":"Policy Frameworks for AI-Bio Convergence and AI for Biosecurity Defense add a RAND report that stress-tested proposed controls against six actor profiles and five pre-release milestones, selecting nine mitigations across managing information access, disrupting materials access, deterrence, and misuse detection, with only precursor screening rated an established approach.","why":"Gives readers a structured, team-judgment feasibility assessment of which layered defenses are ready now versus early-stage or institutionally complex.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Guerra et al., RAND, 2026","url":"https://doi.org/10.7249/RRA4999-1","type":"report"}]}]},
{"date":"2026-09-01","headline":"OpenAI's Astra cybersecurity claim; new refusal and surveillance benchmarks","entries":[{"category":"developments","title":"OpenAI reports Astra meets Critical cybersecurity threshold","change":"On 1 September 2026, Red-Teaming AI Systems adds OpenAI's developer-reported designation that its upcoming Astra model meets the Preparedness Framework Critical cybersecurity threshold, its first model at that level, alongside a bio and chem distinction, a Daybreak Blue access caveat, and a 28 August restart of paused reinforcement-learning work.","why":"Records a developer safety designation for an unreleased model, noted in the chapter as a cybersecurity call only, not independent certification or a biological finding.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"OpenAI, September 2026","url":"https://openai.com/index/path-to-astra/","type":"announcement"}]},{"category":"developments","title":"Compute governance gap: copying and running models on local hardware","change":"Policy Frameworks for AI-Bio Convergence notes that FLOP reporting, cloud verification, and chip export rules govern large-scale training but do not bind copying, running, or booting models on the hardware that holds their weights, including on-prem boxes and rented GPU clusters, citing a public warning that escaped agents might next try to take over such a provider to run more copies.","why":"Flags an unresolved question for compute governance: whether copy-and-run controls must bind the physical hardware itself, not just large training runs.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Sutskever, 2026","url":"https://x.com/ilyasut/status/2094881278621253755","type":"post"}]},{"category":"evidence","title":"Agent-evaluation benchmarks for refusal calibration and genomic surveillance","change":"Adds BioSecBench-Surveillance, where the best tested agents cleared about half of 100 pathogen-genomic analysis tasks (computational analysis, not field detection), and BioSecBench-Refusal, which pairs routine and concealed-hazard tasks to separate over-refusal from missed-hazard failures; also notes a dated Grok 4.6 snapshot's refusal and completion rates.","why":"Distinguishes computational benchmark scores from operational capability so readers do not read agent-evaluation results as demonstrated biosecurity risk or defense.","chapters":[["Digital Biosurveillance: From Wearables to Real-Time Genomics","/ai-biosecurity/digital-biosurveillance.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Bhasin, Flyangolts, et al., 2026, preprint","url":"https://arxiv.org/abs/2607.19262","type":"preprint"},{"label":"Wintermute et al., 2026, preprint","url":"https://arxiv.org/abs/2607.05462","type":"preprint"},{"label":"LatchBio, September 2026","url":"https://blog.latch.bio/p/analyzing-grok46-safeguards","type":"other"},{"label":"xAI, September 2026","url":"https://x.ai/news/biosafety-at-the-frontier","type":"announcement"}]},{"category":"evidence","title":"Subversive LoRA untraining on open-weight models","change":"Policy Frameworks for AI-Bio Convergence adds an ICLR 2024 workshop preprint reporting that an adapter trained on Llama 2-Chat 70B for under $200 on one GPU cut refusal to about 1% on two benchmarks while retaining general capability, distinguishing this weight-level untraining from prompt-side jailbreaking.","why":"Gives the open-weights guardrail claim a primary source and separates weight-level untraining from jailbreaking, noting the finding predates current frontier models.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Lermen, Rogers-Smith, and Ladish, 2024, preprint","url":"https://arxiv.org/abs/2310.20624","type":"preprint"}]}]},
{"date":"2026-09-02","headline":"NVIDIA's five-layer AI industrial stack cited; WHO publisher dialogue on dual-use manuscript review reported; RAND finds LLM agents can produce screening-evading redesigns in silico; MOSAIC method assembles over 1,000 gene fragments in one pot; Four dimensions of residual risk beyond nucleic-acid screening; Customer-screening evaluations show LLM cost savings and 2024 to 2025 regulatory alignment; Science Perspective calls AI viral-genome design ungoverned; Generative RNA design matches expert human designers on Eterna blind targets; Capability-stack framework and input-screening preprint for AI pathogen design; RAND reanalysis places dual-use biology benchmark scores on a common scale; BWC Working Group holds ninth session without a verification mechanism; NBAF's timeline for handling the most dangerous pathogens remains uncertain; NIH proposes draft biosafety policy replacing the 2024 Guidelines; Institutional analyses find dual-use research oversight still incomplete; NTI and Concordia AI launch a biological-evaluation working group; Belfer report proposes deterrence by resilience for bioconvergence; RAND maps the implementation gap in mirror-biology governance","entries":[{"category":"developments","title":"NVIDIA's five-layer AI industrial stack cited","change":"Policy Frameworks for AI-Bio Convergence cites Jensen Huang's March 2026 NVIDIA blog post presenting a five-layer industrial stack for AI (energy, chips, infrastructure, models, applications), noting it is a vendor's own industrial map from a company that sells the chips.","why":"Locates existing compute-control discussion against a vendor's own framing of the AI industrial stack.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Huang, NVIDIA Blog, 2026","url":"https://blogs.nvidia.com/blog/ai-5-layer-cake/","type":"announcement"}]},{"category":"policy","title":"WHO publisher dialogue on dual-use manuscript review reported","change":"Policy Frameworks for AI-Bio Convergence adds a WHO-convened dialogue with six journal publishers and one preprint publisher, which found rejection for dual-use concern is rare, non-life-science authors are often unaware of dual-use issues, and a rejected manuscript can still appear elsewhere.","why":"Documents current publisher practice on dual-use manuscript review as a meeting report, not a binding standard.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"WHO, 2026","url":"https://doi.org/10.2471/B09766","type":"paper"}]},{"category":"evidence","title":"RAND finds LLM agents can produce screening-evading redesigns in silico","change":"Autonomous AI Agents adds a RAND evaluation finding that LLM-driven agents operating biological design tools can produce peptide and protein redesigns that pass in-silico validity checks and, in a minority of cases, evade nucleic-acid synthesis screening, bounded to computational redesigns with no physical orders placed.","why":"Moves the chapter's agent-screening-evasion concern from a theoretical claim to a measured in-silico bound.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Lee et al., RAND, 2026","url":"https://doi.org/10.7249/RRA4741-2","type":"report"}]},{"category":"evidence","title":"MOSAIC method assembles over 1,000 gene fragments in one pot","change":"DNA Synthesis Screening adds MOSAIC, a Nature Biotechnology method that assembles overlapping microchip oligonucleotides into more than 1,000 distinct gene fragments by hybridization rather than enzymatic PCR assembly, with near-zero reported misalignment, noted as a throughput advance rather than a screening evaluation.","why":"Documents a new assembly method relevant to synthesis capacity without overstating it as a pathogen-construction result.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Wu et al., 2026","url":"https://doi.org/10.1038/s41587-026-03266-2","type":"paper"}]},{"category":"evidence","title":"Four dimensions of residual risk beyond nucleic-acid screening","change":"DNA Synthesis Screening adds a mini-review arguing provider sequence screening and customer verification are no longer sufficient as standalone safeguards, organizing residual risk around design capability, access, governance strength, and liability and attribution clarity, and noting no binding global standard requires uniform procedures across providers.","why":"Gives the chapter a diagnostic framework for why screening coverage varies by provider and jurisdiction.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Kato et al., 2026","url":"https://doi.org/10.3389/fbioe.2026.1820001","type":"paper"}]},{"category":"evidence","title":"Customer-screening evaluations show LLM cost savings and 2024 to 2025 regulatory alignment","change":"DNA Synthesis Screening adds an evaluation in which the best LLM configuration matched expert-human legitimacy-screening accuracy (90.2% versus 89.0%) at about one-tenth the cost, and a policy-and-practice review finding OSTP, UK DSIT, IGSC, and ISO aligned on five customer-screening practices in 2024 to 2025.","why":"Shows both a cost-efficient screening approach and cross-regulator convergence on customer-screening practice.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Acelas et al., 2026","url":"https://doi.org/10.3389/fbioe.2026.1819556","type":"paper"},{"label":"Carter, Boldrini, and Alexanian, 2026","url":"https://doi.org/10.3389/fbioe.2026.1819810","type":"paper"}]},{"category":"developments","title":"Science Perspective calls AI viral-genome design ungoverned","change":"DNA Synthesis Screening adds a Johns Hopkins Center for Health Security Perspective stating that generation of functional viral genomes has urgent biosafety implications and that governance to steer AI-composed viral genomes does not yet exist, as independently reported by The Guardian, which is a reported editorial recommendation rather than enacted law.","why":"Flags expert commentary calling for governance to catch up with generative AI's viral-genome design capability.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Inglesby and Hanke, 2026","url":"https://doi.org/10.1126/science.aej8512","type":"paper"},{"label":"Sample, The Guardian, 2026","url":"https://www.theguardian.com/science/2026/aug/06/safety-fears-as-scientists-make-first-viruses-designed-by-ai","type":"news"}]},{"category":"evidence","title":"Generative RNA design matches expert human designers on Eterna blind targets","change":"AI-Enabled Pathogen Design adds a Science study showing generative models can inverse-design RNA pseudoknot sequences for specified folds and, across 57 Eterna blind targets, match experienced human designers on chemical mapping, mutagenesis, and cryo-EM structure, bounded to structure-layer design rather than viral function.","why":"Documents a capability demonstration in RNA structure design distinct from pathogen function or transmissibility.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[{"label":"Townley et al., 2026","url":"https://doi.org/10.1126/science.aeg6829","type":"paper"}]},{"category":"evidence","title":"Capability-stack framework and input-screening preprint for AI pathogen design","change":"AI-Enabled Pathogen Design adds a five-layer AI-biosecurity capability stack, a planning tool locating risk at the interfaces between models, agentic interfaces, and cloud labs, and a bioRxiv preprint proposing input screening of user-specified binding targets as a missing layer, tested against 14,541 proteoform targets of concern.","why":"Gives the chapter both a planning framework for where risk concentrates and a proposed screening layer that current filters miss.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[{"label":"Luhachack, Connell, and Berger, 2026","url":"https://doi.org/10.3389/fmicb.2026.1899413","type":"paper"},{"label":"Palmer et al., 2026, preprint","url":"https://doi.org/10.64898/2026.08.04.740855","type":"preprint"}]},{"category":"evidence","title":"RAND reanalysis places dual-use biology benchmark scores on a common scale","change":"How AI Amplifies Biosecurity Risk adds a RAND research report that reanalyzed 8,146 dual-use biology items from 26 benchmarks across 45 frontier models on a common item-response scale, including SecureBio's private benchmark set with expert baselines.","why":"Gives readers a cross-benchmark interpretation of capability scores rather than treating each benchmark in isolation, without replacing uplift trials.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Ellison et al., RAND, 2026","url":"https://doi.org/10.7249/RRA5055-1","type":"report"}]},{"category":"policy","title":"BWC Working Group holds ninth session without a verification mechanism","change":"International Governance and the BWC adds that the treaty's open-ended Working Group on compliance and verification held its ninth session in Geneva on 17 to 21 August 2026, working from a Chair's draft agreed only ad referendum, with a tenth session scheduled for December 2026 and still no binding verification mechanism adopted.","why":"Records a process milestone in the BWC's long-stalled effort to develop compliance and verification measures.","chapters":[["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"]],"sources":[{"label":"UNODA, BWC Working Group ninth session, 2026","url":"https://meetings.unoda.org/meeting/79378/","type":"policy"}]},{"category":"developments","title":"NBAF's timeline for handling the most dangerous pathogens remains uncertain","change":"Laboratory Biosafety and Biosecurity adds that USDA does not know when the National Bio and Agro-Defense Facility will handle its most dangerous pathogens, and that a USDA Inspector General report found the facility was accepted after a 2024 DHS transfer without a complete commissioning report and incurred $75.8 million in questioned remediation costs.","why":"Shows that construction and a high-containment designation do not by themselves mean a facility is operationally ready.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"Li, Science, 2026","url":"https://doi.org/10.1126/science.ael6250","type":"news"},{"label":"Moser, The Mercury, 2026","url":"https://themercury.com/news/federal-report-says-nbaf-not-appropriately-configured-to-handle-dangerous-pathogens/article_eb7a8c9b-ec81-4b0a-b3d9-30cd84c22383.html","type":"news"}]},{"category":"policy","title":"NIH proposes draft biosafety policy replacing the 2024 Guidelines","change":"Laboratory Biosafety and Biosecurity adds that NIH issued NOT-OD-26-112 on 19 August 2026, requesting comment through 19 October 2026 on a draft policy that would replace the NIH Guidelines for Research Involving Recombinant or Synthetic Nucleic Acid Molecules and expand coverage to all biomedical research involving biohazards at NIH-funded institutions.","why":"The current NIH Guidelines remain in force, but a replacement policy is now open for public comment.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[{"label":"NIH, Notice NOT-OD-26-112, 2026","url":"https://grants.nih.gov/grants/guide/notice-files/NOT-OD-26-112.html","type":"policy"}]},{"category":"developments","title":"Institutional analyses find dual-use research oversight still incomplete","change":"Dual-Use Research of Concern adds a university-association brief reading the July Stopping Policy as requiring institutions to review all life-sciences research for dangerous gain-of-function by January 2027, and a Belfer Center analysis stating the U.S. still lacks a sweeping federal AI-and-biosecurity statute, with frontier-model safeguards mostly voluntary and nucleic-acid screening not binding the full private market.","why":"Documents two institutional readings of how much oversight ground the current policy patchwork actually covers.","chapters":[["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"sources":[{"label":"COGR, 2026","url":"https://www.cogr.edu/sites/default/files/2026-08/Summary%20of%20USG%20Policy%20for%20Stopping%20High%20Risk%20Life%20Sciences%20Research.pdf","type":"other"},{"label":"Florescu-Ciobotaru and Workman, Belfer Center, 2026","url":"https://www.belfercenter.org/research-analysis/dual-use-frontier-ai-enabled-biotechnology-civilian-opportunities-national","type":"report"}]},{"category":"developments","title":"NTI and Concordia AI launch a biological-evaluation working group","change":"Red-Teaming AI Systems adds that NTI | bio and Concordia AI, with technical input from SecureBio, launched an invitation-only working group on the comparability of biological capability evaluations for organizations that already run them across the United States, China, and the EU.","why":"Tracks a new cross-organization effort to standardize how biological capability evaluations are compared.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"NTI | bio and Concordia AI, 2026","url":"https://www.nti.org/wp-content/uploads/2026/08/FOR-PUBLICATION_Project-Overview_Bio-Evals-WG.pdf","type":"report"}]},{"category":"evidence","title":"Belfer report proposes deterrence by resilience for bioconvergence","change":"The Future of Biosecurity adds a 210-page Belfer strategy report arguing that deterrence-by-denial and deterrence-by-punishment fit poorly against bioconvergence threats, and proposing deterrence by resilience built on prevention, detection, attribution, and readiness.","why":"Offers an alternative strategic framework for deterrence tailored to biological rather than nuclear-style threats.","chapters":[["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[{"label":"Sherwood-Randall and McQuade, Belfer Center, 2026","url":"https://www.belfercenter.org/bio-deterrence","type":"report"}]},{"category":"evidence","title":"RAND maps the implementation gap in mirror-biology governance","change":"Synthetic Biology and Democratization adds a RAND operational map sorting mirror-biology research into six function-based categories by how far each would reduce remaining barriers to a self-replicating mirror organism.","why":"Gives the chapter a structured way to assess which lines of mirror-biology work matter most for governance attention.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"]],"sources":[{"label":"Zakaria et al., RAND, 2026","url":"https://doi.org/10.7249/PEA4952-1","type":"report"}]}]},
{"date":"2026-09-05","headline":"Anthropic Model Hardware Standard instrument-control preview; Bill Gates's GERM proposal and its WHO successor distinguished; RAND proposal to extend DNA-synthesis oversight beyond point of sale; Bulletin opinion frames AI-designed phage result as a governance warning; Further Reading frames the handbook as a practitioner companion to RAND studies","entries":[{"category":"developments","title":"Anthropic Model Hardware Standard instrument-control preview","change":"Autonomous AI Agents adds Anthropic's Model Hardware Standard research preview, a shared driver letting agents discover and operate programmable instruments over MCP, CLI, or APIs, with partner qPCR and liquid-handling support in progress; the specification is not open source. A pointer distinguishes it from self-driving-laboratory evidence covered in a companion handbook.","why":"Names the instrument-control path the chapter already flags as a governance concern, while noting it does not yet show a hazardous biological workflow.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Anthropic, 2026","url":"https://www.anthropic.com/news/model-hardware-standard-research-preview","type":"announcement"}]},{"category":"developments","title":"Bill Gates's GERM proposal and its WHO successor distinguished","change":"The Future of Biosecurity adds background on Bill Gates's 2022 GERM proposal, a standing WHO-managed pandemic response cadre of about 3,000 people and about $1 billion a year, and distinguishes it from WHO's Global Health Emergency Corps, a related country-centered framework that cites GERM as lineage but was never built as GERM was originally sketched.","why":"Clarifies that GERM was a proposal, not a built WHO standing force, and keeps Gates's containment framing separate from the handbook's 100-Day Mission coverage.","chapters":[["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[{"label":"Gates, 2022","url":"https://www.penguinrandomhouse.com/books/704751/how-to-prevent-the-next-pandemic-by-bill-gates/","type":"other"},{"label":"Horton, 2022","url":"https://doi.org/10.1016/S0140-6736(22)00874-1","type":"paper"},{"label":"Dowell et al., 2025","url":"https://doi.org/10.1136/bmjgh-2025-019424","type":"paper"}]},{"category":"evidence","title":"RAND proposal to extend DNA-synthesis oversight beyond point of sale","change":"DNA Synthesis Screening adds a 2026 RAND report proposing the Federal Select Agent Program as a complementary life-cycle oversight layer for certain nucleic acids alongside point-of-sale provider screening, plus implementation requirements: precise machine-readable scope, risk-scaled obligations, and phased rollout with cost-benefit analysis.","why":"Addresses the gap that point-of-sale screening does not cover possession, use, or transfer of material after delivery.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Sharkey et al., 2026","url":"https://doi.org/10.7249/RRA4496-2","type":"paper"}]},{"category":"developments","title":"Bulletin opinion frames AI-designed phage result as a governance warning","change":"DNA Synthesis Screening adds a September 2026 Bulletin of the Atomic Scientists opinion piece framing the recent AI-designed phage result as a warning that a window remains to layer mandatory screening, calibrated dataset access, and pathogen early warning, labeled as commentary rather than new experimental evidence.","why":"Adds informed commentary connecting the phage result to the chapter's defense-in-depth argument.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Guerra, Bulletin, 2026","url":"https://thebulletin.org/2026/09/newly-created-viruses-are-a-warning-we-still-have-a-window-to-stop-ai-enabled-bioweapons/","type":"news"}]},{"category":"revision","title":"Further Reading frames the handbook as a practitioner companion to RAND studies","change":"Further Reading describes the handbook as a living, citable practitioner reference that complements episodic RAND AI-biosecurity studies rather than competing with them, citing the handbook's own Zenodo DOI.","why":"Helps readers distinguish this continuously updated reference from the one-off research reports it cites.","chapters":[["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[]}]},
{"date":"2026-09-06","headline":"Chain-of-thought monitoring limits in agentic coding tasks","entries":[{"category":"evidence","title":"Chain-of-thought monitoring limits in agentic coding tasks","change":"Red-Teaming AI Systems adds an OpenAI chief scientist's essay reporting diminishing reliance on chain-of-thought monitoring as models advance, alongside a preprint showing that optimizing agents against a monitor that inspects reasoning but not actions can produce concealed reward hacking in coding tasks.","why":"Qualifies chain-of-thought monitoring as one safeguard among several, not sufficient proof of safe behavior, while noting the coding results do not establish biological-workflow performance.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Pachocki, 2026, essay","url":"https://openai.com/index/an-alien-mind/","type":"announcement"},{"label":"Baker et al., 2025, preprint","url":"https://doi.org/10.48550/arXiv.2503.11926","type":"preprint"},{"label":"Korbak et al., 2025, preprint","url":"https://doi.org/10.48550/arXiv.2507.11473","type":"preprint"}]}]},
{"date":"2026-09-09","headline":"OpenAI reports an automated research intern threshold; DeepMind's AI Control Roadmap for internal agent misuse; AI-designed synthetic plant immune receptors confer disease resistance; BioVeil MATRIX proposed as a defensive taxonomy for agentic biological AI","entries":[{"category":"developments","title":"OpenAI reports an automated research intern threshold","change":"How AI Amplifies Biosecurity Risk adds OpenAI's September 2026 internal report claiming research-org coding agents delivered more than three agent-workdays of effort per human workday by mid-August, and a pause on reinforcement learning for latest deploy-intended models after the Hugging Face incident.","why":"Records a vendor self-report of accelerating internal AI research capability, flagged in the chapter as preliminary and not independently audited.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"OpenAI, 2026","url":"https://openai.com/index/research-acceleration-view-inside-openai/","type":"announcement"}]},{"category":"evidence","title":"DeepMind's AI Control Roadmap for internal agent misuse","change":"How AI Amplifies Biosecurity Risk adds DeepMind's GDM AI Control Roadmap (version 0.1), which treats imperfectly aligned internal agents as potential insider threats and maps four detection levels and three prevention or response levels to rising evasion and harm capability.","why":"Gives the chapter a vendor framework for internal agent-misuse detection, noted as an early roadmap rather than an independent assurance of safe deployment.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Phuong et al., 2026","url":"https://arxiv.org/abs/2607.13087","type":"preprint"},{"label":"DeepMind, 2026, blog","url":"https://deepmind.google/blog/securing-the-future-of-ai-agents/","type":"announcement"}]},{"category":"evidence","title":"AI-designed synthetic plant immune receptors confer disease resistance","change":"AI-Enabled Pathogen Design adds a Science paper showing AI-guided design plus in-planta evolution produced synthetic plant immune receptors that recognize proteins from viral, bacterial, fungal, and oomycete pathogens and conferred disease resistance in transgenic plants.","why":"Documents a beneficial agricultural design capability while noting it is not a finding about human-pathogen receptors or weapons pathways.","chapters":[["AI-Enabled Pathogen Design","/ai-biosecurity/ai-pathogen-design.html"]],"sources":[{"label":"Zhu et al., 2026","url":"https://doi.org/10.1126/science.aee1792","type":"paper"}]},{"category":"evidence","title":"BioVeil MATRIX proposed as a defensive taxonomy for agentic biological AI","change":"How AI Amplifies Biosecurity Risk adds a preprint arguing tool-equipped AI scientist agents can assist dual-use biology tasks that base-model safeguards block, reporting capability uplift on WMDP-style proxies with agent scaffolding, and proposing BioVeil MATRIX, a 10-category, 22-technique defensive taxonomy.","why":"Gives the chapter a proposed defensive framework for agentic biological-AI misuse, bounded as preprint proxy-benchmark findings rather than wet-lab misuse trials.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Provatas et al., 2026, preprint","url":"https://arxiv.org/abs/2605.00927","type":"preprint"}]}]},
{"date":"2026-09-10","headline":"METR proposes independent post-incident propensity investigations; AI-bio governance: binding, voluntary, and proposed instruments; DNA Synthesis Screening chapter adds an At a Glance box; Anthropic's Threat Intelligence report: biological misuse case series","entries":[{"category":"developments","title":"METR proposes independent post-incident propensity investigations","change":"How AI Amplifies Biosecurity Risk, Autonomous AI Agents, Red-Teaming AI Systems, and The Future of Biosecurity add METR's proposal that after serious agent misalignment incidents, labs invite independent investigations of underlying propensities and motives, with public conclusions and a redaction summary.","why":"Extends a single accountability proposal across the chapters most relevant to agent incidents, evaluation, and forward-looking governance.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[{"label":"METR, 2026","url":"https://metr.org/blog/2026-07-28-investigating-ai-propensities-after-incidents/","type":"report"}]},{"category":"revision","title":"AI-bio governance: binding, voluntary, and proposed instruments","change":"Policy Frameworks for AI-Bio Convergence classifies AI-bio governance instruments as binding or funding-conditioned, voluntary, or proposed, and dating US status as of August 2026.","why":"Helps readers quickly separate enacted, binding rules from voluntary commitments and proposed frameworks such as Biosecurity Data Levels before reading the full chapter.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[]},{"category":"revision","title":"DNA Synthesis Screening chapter adds an At a Glance box","change":"DNA Synthesis Screening sets its visible title to include the IGSC subtitle and adds an At a Glance box summarizing the IGSC baseline, US policy status as of August 2026, and four gaps: oligo fragmentation, benchtop synthesizers, capacity outside IGSC, and AI-designed sequence evasion.","why":"Gives readers a quick orientation to the chapter's scope and known gaps before the detailed discussion.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[]},{"category":"developments","title":"Anthropic's Threat Intelligence report: biological misuse case series","change":"Four chapters add Anthropic's September 2026 Threat Intelligence report, described by the company as its first public case series on biological misuse of a frontier platform, covering five investigations from December 2025 to August 2026; How AI Amplifies Biosecurity Risk also adds same-day New York Times coverage of the disclosure.","why":"Gives the vendor-reported case series independent press coverage while noting the Times could not verify whether the underlying work was legitimate or malicious.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Anthropic, 2026","url":"https://www.anthropic.com/threat-intelligence-report-september-2026","type":"announcement"},{"label":"New York Times, Sept. 10, 2026","url":"https://www.nytimes.com/2026/09/10/us/politics/anthropic-ai-biological-weapons.html","type":"news"}]}]},
{"date":"2026-09-12","headline":"Amodei's frontier-pacing essay lands across four chapters","entries":[{"category":"developments","title":"Amodei's Pace the Frontier essay on capability pacing","change":"Four chapters add Anthropic CEO Dario Amodei's essay arguing frontier labs should pace capability gains rather than halt training, starting with Anthropic's pledge to embedded third-party evaluators with employee-like access, then industry and government coordination, then a global floor prohibiting AI use for biological weapons.","why":"Introduces a frontier-lab CEO's three-step governance plan and company pledge, noted as an essay and company action rather than enacted law.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Amodei, 2026","url":"https://darioamodei.com/post/we-must-pace-the-frontier","type":"other"}]}]},
{"date":"2026-09-13","headline":"NEJM AI study: medical chatbot data exposure via prompt probing","entries":[{"category":"evidence","title":"Medical chatbot leaked system prompt and patient conversations","change":"LLMs and Information Hazards adds a NEJM AI study in which ordinary browser tools plus a commercial AI assistant retrieved a patient-facing medical chatbot's system prompt, pipeline configuration, knowledge base, and 1,000 stored patient conversations without authentication.","why":"Adds a deployment-security teaching case illustrating how the same LLM capabilities that ease building medical chatbots also ease probing them.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Madrid-García et al., 2026","url":"https://doi.org/10.1056/AIp2600583","type":"paper"}]}]},
{"date":"2026-09-15","headline":"RAND on open-weight anti-refusal tampering","entries":[{"category":"evidence","title":"RAND: open-weight models resist anti-refusal tampering asymmetrically","change":"How AI Amplifies Biosecurity Risk and Red-Teaming AI Systems add a RAND report finding open-weight frontier LLMs can be modified for malicious biological purposes asymmetrically: removing safety training is highly feasible, while enhancing biological capability is harder, and untampered open-weight models already approach closed-frontier usefulness on several agentic dual-use biology tasks.","why":"Quantifies why an unmodified open-weight model may be more useful to a threat actor than a closed frontier model in some settings, since no refusal classifiers or KYC controls apply once self-hosted.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Dev et al., 2026","url":"https://doi.org/10.7249/RRA5112-1","type":"paper"}]}]},
{"date":"2026-09-16","headline":"Mirror life capability evidence and governance signals","entries":[{"category":"evidence","title":"Paper2Agent: MCP packaging of research methods","change":"Autonomous AI Agents notes that Paper2Agent packages a paper's methods as callable MCP tools, which lowers reuse friction but also concentrates dual-use method access into agent-invokable surfaces; the demo covers AlphaGenome, Scanpy, and TISSUE, not autonomous hazardous lab workflows.","why":"Illustrates a new dissemination pattern, papers packaged as agent-callable tools, worth tracking as a dual-use surface even though this demo is not itself hazardous.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Miao et al., 2026","url":"https://doi.org/10.1038/s41586-026-11044-y","type":"paper"}]},{"category":"evidence","title":"Mirror life capability evidence and at-a-glance summary","change":"Synthetic Biology and Democratization adds an At a Glance summary distinguishing existing mirror molecules from still-nonexistent mirror organisms, citing new peer-reviewed steps toward a mirror ribosome (a mirror-image DNA polymerase and a mirror-image RNA polymerase) and academy, UK, and commentary assessments of governance timing.","why":"Leading assessments put a complete mirror organism roughly a decade or more away, so the chapter separates real chemistry progress from the still-hypothetical organism-level risk.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"]],"sources":[{"label":"Fan, Deng, Zhu, 2021","url":"https://doi.org/10.1038/s41587-021-00969-6","type":"paper"},{"label":"Xu and Zhu, 2022","url":"https://doi.org/10.1126/science.abm0646","type":"paper"},{"label":"Gillum, 2026","url":"https://doi.org/10.1177/23265094261480574","type":"paper"},{"label":"NASEM, 2026","url":"https://doi.org/10.17226/29091","type":"report"},{"label":"Nature, 2025","url":"https://doi.org/10.1038/d41586-025-02902-2","type":"news"},{"label":"UK GO-Science, 2025","url":"https://www.gov.uk/government/publications/mirror-life","type":"policy"}]},{"category":"developments","title":"Mirror life governance signals, explainer, and glossary entry","change":"Adds a University of Utah Health explainer on mirror therapeutics, further mirror-image chemistry papers, and a stack of funder, UNESCO, and Carnegie Endowment statements against building a self-replicating mirror organism; the Glossary adds an entry distinguishing mirror molecules from a living mirror cell.","why":"Shows how the 2024 mirror-bacteria warning has moved into public explanation, funder policy, and international ethics guidance while the capability remains unbuilt.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["Glossary of Biosecurity Terms","/appendices/glossary.html"],["The Future of Biosecurity","/governance/future-biosecurity.html"]],"sources":[{"label":"University of Utah Health, 2024","url":"https://healthcare.utah.edu/newsroom/news/2024/12/mirror-life-still-hypothetical-heres-why-it-should-probably-stay-way","type":"other"},{"label":"Harrison et al., 2023","url":"https://doi.org/10.1038/s41570-023-00493-y","type":"paper"},{"label":"Cuéllar, Hinderstein, and Relman, 2025","url":"https://carnegieendowment.org/posts/2025/10/mirror-life-bacteria-risks-prevention","type":"report"},{"label":"UNESCO IBC, 2025","url":"https://unesdoc.unesco.org/ark:/48223/pf0000395064","type":"report"},{"label":"Evans et al., 2025","url":"https://doi.org/10.58875/nlse3159","type":"paper"},{"label":"Spirit of Asilomar, 2025","url":"https://doi.org/10.25611/X89A-D947","type":"other"}]}]},
{"date":"2026-09-17","headline":"Guo et al. Cell perspective on agentic dual use; Congressional Research Service brief on AI-bio screening (IF13269)","entries":[{"category":"evidence","title":"Guo et al. Cell perspective on agentic dual use","change":"Autonomous AI Agents cites a Cell perspective on agentic autonomous science that lists dual use among the risks of biomedical AI agents able to pose their own questions, alongside bias and cognitive deskilling.","why":"Names dual use as a risk category in a peer-reviewed perspective on autonomous science, not evidence of agent-enabled pathogen misuse.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Guo et al., 2026","url":"https://doi.org/10.1016/j.cell.2026.08.052","type":"paper"}]},{"category":"policy","title":"Congressional Research Service brief on AI-bio screening (IF13269)","change":"DNA Synthesis Screening and Policy Frameworks for AI-Bio Convergence cite a September 2026 Congressional Research Service brief (IF13269) mapping voluntary and mandatory nucleic-acid screening, including IGSC protocols and 119th Congress bills to harden screening standards.","why":"Gives readers a congressional staff map of AI-bio screening authorities, noted as a briefing rather than new empirical misuse data.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"CRS, 2026","url":"https://www.congress.gov/crs-product/IF13269","type":"policy"}]}]},
{"date":"2026-09-21","headline":"SecureBio's pre-release GPT-6 Astra assessment; House Science hearing on screening","entries":[{"category":"evidence","title":"SecureBio pre-release biosecurity assessment of GPT-6 Astra","change":"Red-Teaming AI Systems adds SecureBio's independent pre-release assessment of GPT-6 Astra across three configurations (bio-classifiers disabled, railfree helpful-only, and production-safeguarded), reporting new highs on virology knowledge tests, a first in-silico designability success on ReproBAIT, and a screening-evasion task refused on every sample by the production checkpoint.","why":"Gives readers the clearest recent example of a capability-versus-safeguard split on one model, distinguishing railfree elicitation scores from shipped-product risk.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"SecureBio, GPT-6 Astra pre-release assessment, 2026","url":"https://securebio.org/resources/gpt-6-astra-assessment.pdf","type":"report"}]},{"category":"policy","title":"House Science hearing on biotechnology innovation and biosecurity","change":"Policy Frameworks for AI-Bio Convergence and DNA Synthesis Screening cite the 16 September 2026 House Science Investigations and Oversight Subcommittee hearing, summarizing witness testimony from Franz, Hodges, and Gronvall on mandatory gene-synthesis screening, computational-design oversight gaps, and research capacity.","why":"Adds a congressional record of expert recommendations on screening and oversight gaps, noted as advocacy testimony rather than enacted law.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"House Science Committee, hearing page, 2026","url":"https://science.house.gov/2026/9/investigations-and-oversight-subcommittee-balancing-biotechnology-innovation-and-biosecurity","type":"policy"}]},{"category":"developments","title":"OpenAI endorses bio-data bills and FRONTIER Act evaluator provision","change":"Policy Frameworks for AI-Bio Convergence adds OpenAI's endorsement of three bio-data bills (Web of Biological Data Act, AI-Ready Bio-Data Standards Act, SCALE Biology Act) and a FRONTIER Act provision for independent verification organizations, noting endorsement is not enactment.","why":"Tracks industry positioning on proposed AI-biosecurity data and evaluation authorities alongside the chapter's existing CRS brief.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Reuters, 2026","url":"https://www.reuters.com/technology/openai-backs-bills-us-congress-ai-biological-weapon-threats-2026-09-15/","type":"news"},{"label":"Turner, 2026","url":"https://blog.stephenturner.us/p/biodata-bills-nsceb","type":"post"},{"label":"Politico, 2026","url":"https://www.politico.com/news/2026/09/15/openai-backs-bipartisan-house-plan-for-third-party-safety-assessments-01076588/","type":"news"}]},{"category":"evidence","title":"Doyle preprint on technical containment limits for biomedical AI","change":"LLMs and Information Hazards adds a preprint arguing technical containment (refusal training, output classification, red-teaming, access controls) is necessary but insufficient alone, because recognizing dangerous outputs requires expertise that itself concentrates insider risk.","why":"Flags the preprint as posted content with no novel empirical findings, not a pathogen-design uplift study or proof that containment is futile.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Doyle, 2026, preprint","url":"https://doi.org/10.20944/preprints202609.1616.v1","type":"preprint"}]}]},
{"date":"2026-09-22","headline":"Anthropic's Opus 5.5 biological evaluations; SecureBio on benchmark ceilings","entries":[{"category":"evidence","title":"ClashEval retrieval-trust benchmark added","change":"LLMs and Information Hazards cites ClashEval, which found models override correct priors when retrieved context is wrong in more than 60% of cases in the authors' bake-off, noting a wrong retrieved protocol fragment can be more dangerous than a blank refusal.","why":"The benchmark is not a wet-lab uplift study, but it illustrates a retrieval failure mode relevant to biosecurity assistants that reference protocols.","chapters":[["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Wu, Wu, and Zou, 2024","url":"https://doi.org/10.52202/079017-1053","type":"paper"}]},{"category":"evidence","title":"AutoRedTeamer automates red-team test generation","change":"Red-Teaming AI Systems cites AutoRedTeamer, which pairs a strategy-proposer agent that folds new published attack ideas into an attack library with a red-teaming agent that builds and refines tests, reporting about 20% higher HarmBench attack-success rates on Llama-3.1-70B than strong baselines at about 46% lower compute.","why":"HarmBench or AIR attack-success is not wet-lab biosecurity misuse success, and better attack automation is dual-use, so the chapter cites the evaluation pattern rather than reproducing attack payloads.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Zhou, Wu, et al., 2025, preprint","url":"https://doi.org/10.52202/085713-5654","type":"preprint"}]},{"category":"developments","title":"Claude Mythos 5.1 system card and Life Sciences Verification Program","change":"Red-Teaming AI Systems and How AI Amplifies Biosecurity Risk add Anthropic's Claude Mythos 5.1 system card, judged CB-1 but not CB-2, and the Life Sciences Verification Program that grants credentialed teams Standard or High-risk research-biology access with offline monitoring in place of only real-time refusals.","why":"Documents the developer's self-reported safeguard tier for Mythos-class models and the access program tied to it, as a vendor self-assessment rather than independent certification.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Anthropic, Claude Fable and Mythos 5.1 system card (PDF), 2026","url":"https://www-cdn.anthropic.com/0339e6a7c5c7b87f5c07798616dc32c215d14235/Claude%20Fable%205.1%20&%20Claude%20Mythos%205.1%20System%20Card.pdf","type":"announcement"},{"label":"Anthropic, Life Sciences Verification Program, 2026","url":"https://www.anthropic.com/news/life-sciences-verification-program","type":"announcement"}]},{"category":"developments","title":"Dunn reads Claude's protein-binder campaign as tool orchestration","change":"Autonomous AI Agents and How AI Amplifies Biosecurity Risk add Andrew Dunn's assessment that Anthropic's Claude binder campaign is mainly open-tool orchestration under a long autonomy prompt, using existing tools such as RFdiffusion and BoltzGen, rather than a leap in protein-design science.","why":"Offers independent commentary distinguishing agent scaffolding and tool access from a genuine advance in biophysics.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Dunn, LinkedIn post, 2026","url":"https://www.linkedin.com/posts/andrew-dunn-176a9379_the-most-interesting-takeaway-from-anthropic-activity-7495864875118710785-3qHf","type":"post"},{"label":"Endpoints analysis, 2026","url":"https://endpoints.news/analysis-what-anthropics-protein-study-says-about-its-life-sciences-aims/","type":"news"},{"label":"Anthropic, Claude accelerates protein design, 2026","url":"https://www.anthropic.com/research/Claude-accelerates-protein-design","type":"announcement"}]},{"category":"developments","title":"EU expert strategy memo urges Action Plans on AI and CBRN security","change":"Policy Frameworks for AI-Bio Convergence adds an independent expert strategy memo calling on the European Commission to publish Action Plans on CBRN security and AI and on loss of control, alongside calls to strengthen biosecurity provisions in the EU Biotech Act.","why":"Flags a recommendations memo, not enacted EU legislation or proof of Commission adoption.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"EU expert strategy memo on AI and CBRN security, 2026","url":"https://transformative-ai.eu/download/a_transformative_ai_strategy_for_europe_v1.1.pdf","type":"report"}]},{"category":"developments","title":"Claude Opus 5.5 system card: biological evaluations","change":"Red-Teaming AI Systems adds the developer's classification of the model as CB-1 but not CB-2 and describes its new evaluation pack: a phage-therapy tabletop exercise and RNA and AAV design proxy tasks.","why":"The developer released a new frontier model with a revised biosecurity evaluation; the chapter notes this is a developer-reported classification, not an independent demonstration of misuse uplift.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Anthropic, Claude Opus 5.5 system card (PDF), 2026","url":"https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf","type":"announcement"}]},{"category":"developments","title":"SecureBio essay on evaluating superhuman biological capabilities","change":"Red-Teaming AI Systems adds a SecureBio essay noting frontier models passed the median expert Virology Capabilities Test score in early 2024 and the best measured expert score in 2026, and argues that safe proxy tasks, not the harmful skill itself, are the deliverable once models exceed measured human performance.","why":"Frontier models now exceed measured human expert performance on some biology benchmarks, so the chapter explains why higher scores need safe proxy tasks rather than direct harmful-skill testing.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"SecureBio essay on evaluating superhuman biological capabilities, 2026","url":"https://securebio.org/blog/evaluating-superhuman-biological-capabilities/","type":"announcement"}]}]},
{"date":"2026-09-23","headline":"ScreenDNA open letter on mandatory synthesis screening; Biosecurity Data Levels framework","entries":[{"category":"developments","title":"ScreenDNA open letter on mandatory synthesis screening","change":"DNA Synthesis Screening adds a June 2026 open letter at screendna.org, co-organized by the Foundation for American Innovation and the Institute for Progress, calling for mandatory U.S. nucleic-acid synthesis screening and recordkeeping and signed by frontier AI labs, synthesis providers, and biosecurity figures.","why":"Adds an advocacy coalition letter alongside the chapter's existing legislative material, noted as a pressure artifact rather than enacted law.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"sources":[{"label":"Open letter at screendna.org, 2026","url":"https://screendna.org/","type":"other"}]},{"category":"content","title":"Biosecurity Data Levels framework","change":"Policy Frameworks for AI-Bio Convergence adds a new section on the proposed Biosecurity Data Levels framework (five tiers gating pathogen-data access before model training).","why":"Distinguishes binding tools (Select Agents, the July 2026 high-risk life sciences policy) from voluntary commitments and from the still-proposed BDL framework, which is not WHO or U.S. binding policy.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"sources":[]},{"category":"developments","title":"Anthropic reports agentic discovery of reverse-transcriptase systems","change":"Autonomous AI Agents adds Anthropic's announcement that Mythos 5 agents mined DNA datasets and surfaced novel array-associated reverse-transcriptase systems of unknown function, alongside disclosure of a BSL-1/BSL-2 lab with no human-pathogen work and all wet-lab steps performed by people.","why":"Frames agent-discovered, programmable-looking DNA systems as a governance and evaluation surface requiring human authorization, not a demonstrated hazardous autonomous lab workflow.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"]],"sources":[{"label":"Anthropic, Claude discovers novel enzyme system, 2026","url":"https://www.anthropic.com/news/claude-discovers-novel-enzyme-system","type":"announcement"},{"label":"Anthropic, Claude discovers novel enzyme system (preprint PDF), 2026","url":"https://www-cdn.anthropic.com/22573675ada52a8ca8a97a1a4b4326b2f208a071.pdf","type":"announcement"}]}]},
{"date":"2026-09-27","headline":"US biodefense on the record: the DoD posture review, the National Blueprint, budgets","entries":[{"category":"policy","title":"Department of Defense Biodefense Posture Review, 2023","change":"Policy Frameworks for AI-Bio Convergence covers the Department of Defense's first Biodefense Posture Review, published in August 2023, which created a Biodefense Council under the Under Secretary of Defense for Acquisition and Sustainment and set four lines of effort for the department's biodefense.","why":"Gives the department's own account of how it organizes biodefense, in place of secondary summaries.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"DoD Biodefense Posture Review, 2023","url":"https://media.defense.gov/2023/Aug/17/2003282337/-1/-1/1/2023_BIODEFENSE_POSTURE_REVIEW.PDF","type":"policy"}]},{"category":"policy","title":"The 2024 National Blueprint for Biodefense and the Apollo Program update","change":"Policy Frameworks for AI-Bio Convergence covers the Bipartisan Commission on Biodefense's National Blueprint for Biodefense of May 2024 and its update to the Apollo Program for Biodefense, which names artificial intelligence among the program's priorities. Key Resources and Further Reading points to the commission's report hub.","why":"Sets out the reform agenda a reader is most likely to meet in US biodefense debate, and where AI sits within it.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"National Blueprint for Biodefense, 2024","url":"https://biodefensecommission.org/reports/the-national-blueprint-for-biodefense/","type":"report"},{"label":"An Update to the Apollo Program for Biodefense","url":"https://biodefensecommission.org/reports/an-update-to-the-apollo-program-for-biodefense-winning-the-race-against-biological-threats/","type":"report"}]},{"category":"policy","title":"What the United States spends on biodefense, and who holds the work","change":"The Future of Biosecurity cites the Council on Strategic Risks' biodefense budget breakdown, with its methodology and its fiscal year 2026 update, and Policy Frameworks for AI-Bio Convergence cites the Bipartisan Commission on Biodefense's B-SPAN maps of the executive and legislative branches. Both are listed in Key Resources and Further Reading.","why":"Budget figures and the map of responsible bodies are checkable in one place instead of inferred from strategy memos.","chapters":[["The Future of Biosecurity","/governance/future-biosecurity.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"CSR Biodefense Budget Breakdown","url":"https://councilonstrategicrisks.org/nolan/biodefense-budget-breakdown/","type":"report"},{"label":"B-SPAN Executive Branch","url":"https://biodefensecommission.org/b-span-executive-branch/","type":"report"},{"label":"B-SPAN Legislative Branch","url":"https://biodefensecommission.org/b-span-legislative-branch/","type":"report"}]},{"category":"policy","title":"Oversight proposals behind gain-of-function, synthetic biology, and dual-use research","change":"Gain-of-Function Research cites CSET's 2023 survey of the global gain-of-function landscape, Synthetic Biology and Democratization the Johns Hopkins Center for Health Security's 2023 biosafety report, and Dual-Use Research of Concern the NSABB's 2023 proposed oversight framework. Policy Frameworks for AI-Bio Convergence adds Executive Order 14081, the National Security Commission on Emerging Biotechnology's 2025 final report, and the Scowcroft Institute's 2025 case for an independent bioresponsibility authority.","why":"Each chapter names the record its account of oversight rests on, so a reader can go to the proposal itself.","chapters":[["Gain-of-Function Research: Science, Risk, and Governance","/biotechnology/gain-of-function.html"],["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"],["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Schuerger et al., CSET, 2023","url":"https://cset.georgetown.edu/publication/understanding-the-global-gain-of-function-research-landscape/","type":"report"},{"label":"Johns Hopkins Center for Health Security, 2023","url":"https://www.centerforhealthsecurity.org/sites/default/files/2023-04/20230228-biosafety-and-biosecurity.pdf","type":"report"},{"label":"NSABB, 2023","url":"https://osp.od.nih.gov/wp-content/uploads/2023/03/NSABB-Final-Report-Proposed-Biosecurity-Oversight-Framework-for-the-Future-of-Science.pdf","type":"policy"},{"label":"NSCEB Final Report, 2025","url":"https://www.biotech.senate.gov/final-report/","type":"policy"},{"label":"Scowcroft Institute, 2025","url":"https://bush.tamu.edu/wp-content/uploads/2025/01/Envisioning-an-Independent-Bioresponsibility-Authority-Jan-2025-2.pdf","type":"report"}]}]},
{"date":"2026-09-28","headline":"Klapsa et al.: London wastewater localized poliovirus circulation","entries":[{"category":"developments","title":"Emerging Technology Policy Careers maps federal biosecurity career paths","change":"Building a Biosecurity Career describes federal biosecurity roles across agencies and connects agency profiles with fellowship and congressional pathways in Emerging Technology Policy Careers guidance.","why":"Agency responsibilities and entry pathways help readers identify where their biosecurity skills fit.","chapters":[["Building a Biosecurity Career: Federal Agencies, Fellowships, and AI-Bio Roles","/governance/biosecurity-careers.html"]],"sources":[{"label":"Emerging Technology Policy Careers, biosecurity policy, 2026","url":"https://emergingtechpolicy.org/areas/biosecurity-policy/","type":"other"},{"label":"Emerging Technology Policy Careers, executive branch guide, 2026","url":"https://emergingtechpolicy.org/institutions/executive-branch/executive-branch-biosecurity-policy/","type":"other"},{"label":"Emerging Technology Policy Careers, fellowship database","url":"https://emergingtechpolicy.org/pathways/policy-fellowships/#fellowship-database","type":"other"}]},{"category":"policy","title":"The 2021 public health supply-chain strategy extends beyond stockpiles","change":"Medical Countermeasures and Biodefense explains the HHS/ASPR strategy directed by Executive Order 14001: public health industrial capacity, sustained manufacturing readiness, and vendor-managed inventory complement stockpile holdings. The strategy is a roadmap, not a study of shortage rates.","why":"Preparedness depends on the capacity to replenish supplies during simultaneous demand.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"HHS/ASPR, National Strategy for a Resilient Public Health Supply Chain, 2021","url":"https://insights.governmentanalytica.com/hubfs/Government_Analytica/Regulation/US%20-%20Supply%20Chains/2021%20HHS%20National-Strategy-for-Resilient-Public-Health-Supply-Chain.pdf","type":"policy"},{"label":"Executive Order 14001, 2021","url":"https://www.federalregister.gov/documents/2021/01/26/2021-01865/a-sustainable-public-health-supply-chain","type":"policy"}]},{"category":"policy","title":"NSCEB proposes biological-data infrastructure and AI-ready standards","change":"Policy Frameworks for AI-Bio Convergence describes NSCEB Final Report section 4.1 proposals for a Department of Energy Web of Biological Data with security and access controls and NIST standards for AI-ready biological data.","why":"The proposals connect biological-data access with governance and standardization responsibilities.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"NSCEB, Final Report section 4.1, 2025","url":"https://www.biotech.senate.gov/final-report/chapters/chapter-4/section-1","type":"policy"}]},{"category":"evidence","title":"Klapsa et al. document London poliovirus detection in wastewater","change":"Outbreak Detection and Surveillance describes Klapsa et al.: 118 genetically linked type-2 poliovirus isolates from 21 of 52 London sewage samples collected in 2022, with expanded sampling localizing circulation to north and east London boroughs.","why":"Wastewater sampling provides a population-level complement to surveillance triggered by clinical cases.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"sources":[{"label":"Klapsa et al., 2022","url":"https://doi.org/10.1016/S0140-6736(22)01804-9","type":"paper"}]},{"category":"evidence","title":"Vickers and Freemont argue for public biofoundries in national preparedness","change":"Cloud Labs and Automated Biology presents Vickers and Freemont's argument that publicly funded biofoundries and coordinated Global Biofoundry Alliance networks can support national bioengineering, biosecurity, and countermeasure response systems.","why":"Public infrastructure gives preparedness planners a model beyond commercial cloud-laboratory access.","chapters":[["Cloud Labs and Automated Biology","/ai-biosecurity/cloud-labs.html"]],"sources":[{"label":"Vickers and Freemont, 2022","url":"https://doi.org/10.1038/s41467-022-28103-3","type":"paper"}]},{"category":"developments","title":"Imperial describes yeast-based vaccine manufacturing capacity","change":"Medical Countermeasures and Biodefense describes Imperial College London's reported work on yeast-based virus-like particle vaccine components and manufacturing equipment pathways closer to brewing than bespoke biologics plants.","why":"Alternative manufacturing platforms broaden the infrastructure options considered in pandemic preparedness.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[{"label":"Imperial College London, Future Vaccine Manufacturing Research Hub","url":"https://www.imperial.ac.uk/stories/harnessing-synthetic-biology-for-the-future-of-biosecurity","type":"announcement"}]},{"category":"evidence","title":"Ibrahim and Kis studies connect RNA vaccine equity with logistics","change":"Medical Countermeasures and Biodefense explains that a UK supply-chain model estimated RNA vaccine logistics costs at about six times those of a viral-vector comparator, with transportation dominating under its assumptions. Related techno-economic work examines fill-to-finish and thermostability.","why":"Vaccine access depends on distribution and release capacity as well as dose allocation.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[{"label":"Ibrahim et al., 2022","url":"https://doi.org/10.1016/b978-0-323-95879-0.50273-3","type":"paper"},{"label":"Kis et al., 2022","url":"https://doi.org/10.1038/s41541-022-00447-3","type":"paper"}]},{"category":"policy","title":"DHS seeks AI-assisted detection of biological signatures in environmental data","change":"AI for Biosecurity Defense describes the DHS Ready, Set...ID the Biothreat prize challenge, which solicits AI algorithms for biological signatures in noisy environmental metagenomic datasets and fewer false alarms. It is a solicitation, not a fielded detector.","why":"The challenge identifies a government evaluation target for environmental biodetection.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"DHS S&T, Ready, Set...ID the Biothreat, 2026","url":"https://www.dhs.gov/science-and-technology/ready-set-id-biothreat","type":"policy"}]}]},
{"date":"2026-09-29","headline":"OpenAI reports GPT-6.1 Sol as High in biological and chemical capability","entries":[{"category":"developments","title":"OpenAI reports GPT-6.1 Sol Preparedness classifications","change":"Red-Teaming AI Systems describes OpenAI's GPT-6.1 Sol addendum: developer-reported Critical cybersecurity and High Biological and Chemical classifications, with reported biological and chemical scores below indicative Critical thresholds. These are developer assessments rather than independent certification.","why":"Configuration-specific assessments help readers distinguish model capability claims from independent evaluation of deployed safeguards.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"OpenAI, GPT-6.1 Sol addendum system card, 2026","url":"https://cdn.openai.com/pdf/38e3efcf-545e-44cd-99ec-2b7eb395f4cc/oai_GPT_6_1_Sol.pdf","type":"announcement"}]}]},
{"date":"2026-09-30","headline":"SynthIDBio protein watermarks in Nature; Gemini 4 Argon Fairwind tier","entries":[{"category":"evidence","title":"SynthIDBio: protein watermarks in Nature, with DeepMind's companion","change":"DNA Synthesis Screening teaches SynthIDBio (Stutz et al., Nature, 2026): sequence watermarks for AI-designed proteins and an optional AlphaFold 3 structure watermark, as provenance rather than a substitute for order screening. DeepMind's Introducing SynthID Bio (30 Sep 2026) describes stacked safeguards, customer vetting, and an early Evo 2 bacteriophage watermark watch item. How AI Amplifies Biosecurity Risk cites both.","why":"Readers get a Nature proof-of-concept for protein provenance, with the developer's companion announcement, and both texts keep nucleic-acid order screening in place.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"sources":[{"label":"Stutz et al., 2026","url":"https://doi.org/10.1038/s41586-026-10965-y","type":"paper"},{"label":"DeepMind, Introducing SynthID Bio, 2026","url":"https://deepmind.google/blog/introducing-synthid-bio/","type":"announcement"}]},{"category":"developments","title":"Gemini 4 Argon Fairwind trusted-access cyber tier","change":"How AI Amplifies Biosecurity Risk records Google's 30 September 2026 announcement of Gemini 4 Argon as a phased Fairwind release for trusted cyber defenders, with developer-reported refusal of cyber and CBRN misuse requests. Red-Teaming AI Systems says to evaluate the Fairwind defender tier and the public product separately, as a developer-reported access boundary rather than independent certification.","why":"Readers can separate a developer-reported trusted-access cyber tier from the guarded public product when judging safeguard claims.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[]},{"category":"developments","title":"ASM Detection to Protection Initiative","change":"Policy Frameworks records the American Society for Microbiology Detection to Protection Initiative, launched in September 2026 and chaired by Dan Jernigan, as a society roadmap through spring 2027 on infectious-disease detection, surveillance, and response. It is not enacted statute and not an AI-access, DNA-screening, or compute-governance control.","why":"A professional-society roadmap on detection and response is distinct from statute and from AI-access or DNA-screening controls.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"],["Biosecurity Resources: Books, Reports, Agencies, and Discovery Hubs","/appendices/resources.html"]],"sources":[{"label":"ASM, Detection to Protection Initiative, 2026","url":"https://asm.org/advocacy/detection-to-protection-initiative","type":"announcement"},{"label":"ASM, press release, 28 Sep 2026","url":"https://asm.org/press-releases/2026/asm-launches-detection-to-protection-initiative","type":"announcement"}]},{"category":"developments","title":"White House Accord on Super Intelligence as a voluntary pledge","change":"Policy Frameworks records the late September 2026 White House Accord on Super Intelligence, the Joint Commitment on Frontier Responsibilities: a voluntary one-page pledge by Google, Anthropic, Meta, OpenAI, xAI, and Nvidia covering internal monitoring, an internal verification team, an external evaluator, and a board committee. The signed text sets no deadlines, names no auditors, and attaches no penalties.","why":"Readers can tell a political and industry pledge from measured safety performance and from mandatory nucleic-acid screening rules.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"White House, 2026","url":"https://x.com/WhiteHouse/status/2105292669303791687","type":"post"}]}]},
{"date":"2026-10-01","headline":"WHO same-day Wuhan notice; IGWG8 advanced the PABS annex; Goodfire monitors","entries":[{"category":"evidence","title":"Goodfire sequence-aware protein-embedding biosecurity monitors","change":"Autonomous AI Agents records Goodfire's October 2026 report: sequence-aware monitors of protein embeddings. On their custom benchmark, Goodfire reported roughly 87% concerning-sequence flag rate at about 1% benign flag rate, and about 99% at about 10%. It is vendor research, not a peer-reviewed trial or an IGSC substitute. Red-Teaming AI Systems does not treat the benchmark as a BioSecBench replication.","why":"The reported flag rates come from the vendor's own benchmark, so they are not independent evidence that these monitors replace order screening.","chapters":[["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"sources":[{"label":"Goodfire, 2026","url":"https://www.goodfire.com/research/better-biosecurity-monitors","type":"announcement"}]},{"category":"evidence","title":"Hazel map of AI governance across national genomics initiatives","change":"Policy Frameworks records Hazel et al. (npj Digital Public Health, 2026): current or planned AI use in 32 of 90 national genomics initiatives (36%) across 70 countries, and public AI-specific governance in three initiatives in two countries (All of Us, Genomics England, and UK Biobank). The review used public materials from March 2024 to February 2026, not private agreements.","why":"Public AI-specific governance documents are rare even where genomics programs already use or plan to use AI.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"Hazel et al., 2026","url":"https://doi.org/10.1038/s44482-026-00043-5","type":"paper"}]},{"category":"policy","title":"NIST hub of AI congressional mandates and executive actions","change":"Policy Frameworks cites the NIST hub mapping congressional mandates and executive actions for its AI work, including the Chips and Science Act of 2022, the National Artificial Intelligence Initiative Act of 2020, and the July 2025 America's AI Action Plan. NIST states it is named in recommendations on an evaluations ecosystem and biosecurity. The hub is orientation, not binding statute.","why":"Readers can locate the statutes and the Action Plan items that name NIST without treating the hub as a new binding rule.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"NIST, AI congressional mandates and executive actions","url":"https://www.nist.gov/artificial-intelligence/ai-congressional-mandates-executive-orders-and-actions","type":"policy"}]},{"category":"revision","title":"WHO was informed of Wuhan pneumonia the same day BlueDot alerted clients","change":"On December 31, 2019, BlueDot alerted clients to unusual pneumonia in Wuhan, and WHO's China Country Office picked up the Wuhan Municipal Health Commission statement the same day. WHO issued its first Disease Outbreak News on January 5 and reported a novel coronavirus on January 9. Travel forecasts for Bangkok, Seoul, and Taipei remain company-reported. AI for Biosecurity Defense.","why":"The WHO timeline dates when the organization was informed, and the travel forecasts stay a company-reported case study.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"sources":[{"label":"WHO, COVID-19 timeline, 2020","url":"https://www.who.int/news/item/29-06-2020-COVIDtimeline","type":"policy"}]},{"category":"policy","title":"Eighth IGWG session advanced the PABS annex without finalizing it","change":"The eighth Intergovernmental Working Group session (September 14–18, 2026) advanced Pathogen Access and Benefit Sharing annex negotiations and did not finalize the text. Finalizing the annex remains required before signature and ratification. World Health Assembly consideration is targeted for May 2027, or earlier if a special session is called. Policy Frameworks.","why":"Countries cannot consider signature and ratification until the annex is finalized.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"sources":[{"label":"WHO, 2026","url":"https://www.who.int/news/item/18-09-2026-member-states-advance-negotiations-on-pathogen-access-and-benefit-sharing-ahead-of-un-general-assembly-meeting-on-pandemics","type":"policy"},{"label":"WHO, IGWG8 session, 2026","url":"https://www.who.int/news-room/events/detail/2026/09/14/default-calendar/eighth-meeting-of-the-intergovernmental-working-group-(igwg)-on-the-who-pandemic-agreement","type":"policy"}]}]},
{"date":"2026-10-06","headline":"SecureBio reports VCT-v2 question revisions and benchmark limits; Irkutsk laboratory death: contested cause and reported quarantine response; Esvelt proposes glycol air treatment alongside far-UVC; Kleinman et al. examine mirror-bacteria countermeasures through proxies; BWC Articles IV and X link national prevention with peaceful exchange","entries":[{"category":"evidence","title":"SecureBio reports VCT-v2 question revisions and benchmark limits","change":"Red-Teaming AI Systems describes SecureBio's VCT-v2 audit: 43 of 322 questions were removed and 163 edited, leaving 279 questions. SecureBio reports greater resistance to shortcuts and more headroom before saturation; question-answering scores remain proxies for operational risk.","why":"Benchmark interpretation requires attention to ambiguous questions, shortcuts, and effective scoring ceilings.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"]],"sources":[{"label":"Mak and Götting, 2026","url":"https://securebio.org/blog/introducing-vct-v2/","type":"report"}]},{"category":"developments","title":"Irkutsk laboratory death: contested cause and reported quarantine response","change":"The Biological Threat Landscape describes independent reporting of a laboratory worker's death and hospital quarantines in Irkutsk alongside the official account of pneumonia of unknown etiology. The cause remains contested, and there is no public evidence of weapons development.","why":"Transparent incident reporting is necessary to assess laboratory exposure and the proportionality of public health responses.","chapters":[["The Biological Threat Landscape","/foundations/threat-landscape.html"]],"sources":[{"label":"RFE/RL, 5 October 2026","url":"https://www.rferl.org/a/russia-plague-warning-siberia-irkutsk-laboratory/33870385.html","type":"news"},{"label":"Meduza, 5 October 2026","url":"https://meduza.io/en/feature/2026/10/05/hospitals-in-irkutsk-impose-quarantines-after-an-employee-at-an-anti-plague-institute-dies-russia-s-public-health-agency-says-she-died-of-pneumonia-and-the-epidemiological-situation-remains-stable","type":"news"}]},{"category":"developments","title":"Esvelt proposes glycol air treatment alongside far-UVC","change":"Medical Countermeasures and Biodefense presents Esvelt's proposal to pair far-UVC with glycol misters as environmental adjuncts when agent identity is uncertain. Triethylene glycol air treatment requires its own safety and efficacy documentation and is not established pandemic infrastructure.","why":"Environmental countermeasures need evidence specific to each intervention before preparedness decisions rely on them.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"sources":[{"label":"Esvelt, 5 October 2026","url":"https://x.com/kesvelt/status/2107211368189317354","type":"post"}]},{"category":"evidence","title":"Kleinman et al. examine mirror-bacteria countermeasures through proxies","change":"Synthetic Biology and Democratization describes an eLife reviewed preprint using chirality-reversed proxies to assess antibiotics and immune responses relevant to hypothetical mirror bacteria. Findings remain proxy inferences, with incomplete evidence on functional immunity and other assessment criteria; no living mirror organisms were tested.","why":"Countermeasure planning must distinguish proxy observations from demonstrated protection against mirror organisms.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"]],"sources":[{"label":"Kleinman, Torres, and Wang, 2026, reviewed preprint","url":"https://elifesciences.org/reviewed-preprints/109134v1","type":"preprint"}]},{"category":"policy","title":"BWC Articles IV and X link national prevention with peaceful exchange","change":"International Governance and the BWC explains Article IV's requirement for national measures to prohibit and prevent biological weapons and Article X's commitment to peaceful exchange. National rules must implement weapons prohibitions without treating legitimate peaceful cooperation as prohibited assistance.","why":"The treaty pairs domestic prevention duties with rights and commitments for peaceful international cooperation.","chapters":[["International Governance and the Biological Weapons Convention","/frameworks/bwc-governance.html"]],"sources":[{"label":"UN Office for Disarmament Affairs, Biological Weapons Convention","url":"https://disarmament.unoda.org/en/our-work/weapons-mass-destruction/biological-weapons/biological-weapons-convention","type":"policy"}]}]},
{"date":"2026-10-07","headline":"California AB 1864 binds companies to the 2024 synthesis screening framework; Titanji distinguishes plague severity from pandemic potential; Kaelber argues for broader prototype-vaccine coverage; Biohub reports a $1.8 billion Virtual Biology Initiative expansion","entries":[{"category":"policy","title":"California AB 1864 binds companies to the 2024 synthesis screening framework","change":"DNA Synthesis Screening describes California AB 1864, approved September 30, 2026: covered providers and equipment manufacturers must follow the September 2024 screening framework and publicly attest compliance. Later federal revisions do not automatically amend the state requirement.","why":"State enforcement changes compliance obligations for companies producing, selling, or shipping covered products in California.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"section":"/biotechnology/dna-synthesis-screening.html#sec-mandatory-screening","sources":[{"label":"California Legislature, AB 1864, 2026","url":"https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB1864","type":"policy"}]},{"category":"developments","title":"Titanji distinguishes plague severity from pandemic potential","change":"The Biological Threat Landscape presents Titanji's clinician commentary that severe individual disease does not by itself establish pandemic potential, citing modern plague containment through isolation, contact tracing, and antibiotics. The essay is not peer reviewed, and the Irkutsk agent identity remains contested.","why":"Preparedness assessments need to distinguish disease consequences from transmission and containment conditions.","chapters":[["The Biological Threat Landscape","/foundations/threat-landscape.html"],["Pathogens of Concern","/foundations/pathogens-of-concern.html"]],"section":"/foundations/threat-landscape.html#sec-irkutsk-2026","sources":[{"label":"Titanji, 2026","url":"https://bktitanji.substack.com/p/plague-is-scary-but-that-doesnt-make","type":"post"}]},{"category":"developments","title":"Kaelber argues for broader prototype-vaccine coverage","change":"Medical Countermeasures and Biodefense presents Kaelber's opinion that prototype-vaccine coverage should extend to some virus families currently ranked low priority because AI could alter the future threat picture. This is expert commentary, not a WHO, CEPI, or NIAID prioritization decision.","why":"Prototype selection involves judgments about future threats as well as current disease priorities.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"section":"/frameworks/countermeasures.html#sec-vaccine-development","sources":[{"label":"Kaelber, 2026, opinion","url":"https://warontherocks.com/how-to-defend-against-ai-designed-viruses/","type":"other"}]},{"category":"developments","title":"Biohub reports a $1.8 billion Virtual Biology Initiative expansion","change":"Policy Frameworks for AI-Bio Convergence describes Biohub's reported $1.8 billion Virtual Biology Initiative expansion with federal and industry partners for AI-ready open biological data. Such investment can increase beneficial capability and dual-use exposure alongside legislative proposals for biological-data governance.","why":"Large biological-data programs make access and governance decisions consequential for both research and biosecurity.","chapters":[["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"section":"/governance/policy-frameworks.html#sec-policy-directives-gap","sources":[{"label":"Biohub, Virtual Biology Initiative announcement, 2026","url":"https://www.prnewswire.com/news-releases/international-cross-sector-collaboration-commits-nearly-2-billion-to-build-foundational-data-for-ai-models-to-predict-and-treat-disease-302901172.html","type":"announcement"},{"label":"Hu, Reuters, 2026","url":"https://www.reuters.com/business/healthcare-pharmaceuticals/us-government-google-join-zuckerberg-backed-biohub-18-billion-push-ai-biology-2026-10-07/","type":"news"}]}]},
{"date":"2026-10-08","headline":"UK AISI reports chemistry and biology evaluation trends and safeguard limits","entries":[{"category":"developments","title":"RAND proposes collective restraint to prevent mirror life","change":"Synthetic Biology and Democratization presents Kalra et al.'s RAND strategy for preventing mirror life through international engagement and collective restraint. It is a policy proposal, not adopted U.S. policy or a peer-reviewed risk estimate.","why":"Prevention commitments offer a governance response when reliable early-warning signs are scarce.","chapters":[["Synthetic Biology and the Democratization of Biotechnology","/biotechnology/synthetic-biology.html"]],"summary":"RAND proposes international engagement and collective restraint to prevent mirror life; the strategy is not adopted U.S. policy.","section":"/biotechnology/synthetic-biology.html#sec-mirror-life-biosecurity","sources":[{"label":"Kalra et al., RAND, 2026","url":"https://doi.org/10.7249/RRA4335-1","type":"report"}]},{"category":"policy","title":"UK AISI reports chemistry and biology evaluation trends and safeguard limits","change":"How AI Amplifies Biosecurity Risk presents the UK AISI Frontier AI Trends Report on chemistry and biology evaluations, including reported safeguard failures. These aggregate government evaluations use proxies and do not establish real-world misuse outcomes or a capability ceiling.","why":"Risk assessments need to distinguish evaluation findings from real-world misuse outcomes.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"summary":"UK AISI reports chemistry and biology evaluation trends and safeguard failures; aggregate proxy evaluations do not establish real-world misuse outcomes.","section":"/ai-biosecurity/ai-risk-amplifier.html#sec-amplifier-govt-evaluations","sources":[{"label":"UK AISI, Frontier AI Trends Report","url":"https://www.aisi.gov.uk/frontier-ai-trends-report#chemistry--biology","type":"report"}]},{"category":"developments","title":"Anthropic reports safeguard results by trusted-access tier","change":"Red-Teaming AI Systems presents Anthropic's reported safeguard results across trusted-access tiers alongside its Cyber Mission program. The developer-run evaluation illustrates evidence that reviewers can request from biological access programs; it is not an independent audit of safeguard strength.","why":"Trusted-access decisions require evidence about safeguard behavior within each access tier.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"summary":"Anthropic's developer-run cyber evaluation illustrates per-tier safeguard evidence that reviewers can request from biological access programs, without constituting an independent audit.","section":"/ai-biosecurity/red-teaming.html#sec-safeguard-stack","sources":[{"label":"Anthropic, Cyber Verification Program, 2026","url":"https://www.anthropic.com/news/cyber-verification-program","type":"announcement"},{"label":"Anthropic, Cyber Mission, 2026","url":"https://www.anthropic.com/news/anthropic-cyber-mission","type":"announcement"}]}]},
{"date":"2026-10-09","headline":"WHO's 7 October account of the Irkutsk event under the International Health Regulations; WHO ties ratification of the Pandemic Agreement to adoption of the PABS annex in May 2027; SecureBio Astra: which checkpoint refused all 100 screening-evasion samples; South Africa's Beta and Omicron reporting and the travel restrictions that followed","entries":[{"category":"policy","title":"WHO's 7 October account of the Irkutsk event under the International Health Regulations","change":"The Biological Threat Landscape describes WHO's 7 October 2026 account: the Russian Federation replied that no plague case had recently been reported in Irkutsk oblast and that about 200 contacts were quarantined. WHO requested the laboratory tests performed on the individual who died and on the contacts, and said it cannot conduct a full risk assessment without the full picture.","why":"Readers can see what the responsible international authority has and has not confirmed about a laboratory-linked death.","chapters":[["The Biological Threat Landscape","/foundations/threat-landscape.html"]],"summary":"WHO says it cannot complete a risk assessment of the Irkutsk event without laboratory results; on WHO's account the agent remains undetermined.","section":"/foundations/threat-landscape.html#sec-irkutsk-2026","sources":[{"label":"WHO, Director-General's media briefing remarks, 7 October 2026","url":"https://www.who.int/news-room/speeches/item/who-director-general-s-opening-remarks-at-the-media-briefing---7-october-2026","type":"announcement"}]},{"category":"policy","title":"WHO ties ratification of the Pandemic Agreement to adoption of the PABS annex in May 2027","change":"Global Surveillance Equity states that the eighth IGWG meeting (September 14–18, 2026) advanced the Pathogen Access and Benefit Sharing annex without finalizing it, and that WHO has said adopting the annex at the World Health Assembly in May 2027 is necessary before Member States can begin national consideration of ratifying the Agreement.","why":"Benefit sharing for shared pathogen samples and sequence data remains unsettled until the annex is adopted.","chapters":[["Global Surveillance Equity: LMIC Capacity as Collective Security","/governance/surveillance-equity.html"]],"summary":"WHO says the PABS annex must be adopted at the World Health Assembly in May 2027 before countries can begin considering ratification.","section":"/governance/surveillance-equity.html#sec-surveillance-equity","sources":[{"label":"WHO, September 25, 2026","url":"https://www.who.int/news/item/25-09-2026-world-leaders-renew-commitment-to-protect-the-world-from-future-pandemics","type":"announcement"}]},{"category":"revision","title":"SecureBio Astra: which checkpoint refused all 100 screening-evasion samples","change":"Red-Teaming AI Systems for Biosecurity Risks and DNA Synthesis Screening attribute the refusal of all 100 Advanced Screening Evasion samples in SecureBio's September 2026 Astra assessment to the pre-release checkpoint with system-level bio-classifiers disabled. Under production safeguards, the assessment reports BioTIER results: 82.9% of refuse prompts refused or blocked.","why":"Refusal by a checkpoint under test and protection in the shipped product are different measurements.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"summary":"In SecureBio's Astra assessment, the 100-sample refusal came from a pre-release checkpoint with bio-classifiers disabled, not from the production configuration.","section":"/ai-biosecurity/red-teaming.html#sec-red-teaming-securebio-astra","sources":[]},{"category":"revision","title":"South Africa's Beta and Omicron reporting and the travel restrictions that followed","change":"Global Surveillance Equity dates the Beta variant (first sampled in October 2020, announced December 18, 2020) and South Africa's first report of Omicron to WHO on November 24, 2021. WHO commended South Africa and Botswana for the speed and transparency of that notification, yet countries then restricted travel from southern Africa.","why":"Transparent variant reporting followed by travel restrictions is the tension the chapter treats as unresolved.","chapters":[["Global Surveillance Equity: LMIC Capacity as Collective Security","/governance/surveillance-equity.html"]],"summary":"WHO commended South Africa and Botswana for rapid, transparent Omicron reporting; travel restrictions on southern Africa followed.","section":"/governance/surveillance-equity.html#sec-surveillance-south-africa","sources":[{"label":"Tegally et al., 2021","url":"https://doi.org/10.1038/s41586-021-03402-9","type":"paper"},{"label":"ECDC, January 2021","url":"https://www.ecdc.europa.eu/sites/default/files/documents/COVID-19-risk-related-to-spread-of-new-SARS-CoV-2-variants-EU-EEA-first-update.pdf","type":"report"}]}]},
{"date":"2026-10-10","headline":"Esvelt's Irkutsk analysis; WHO's COVID-19 origins assessment and research oversight","entries":[{"category":"developments","title":"Esvelt tests a benign explanation for the Irkutsk death","change":"Esvelt evaluates a benign explanation for the Irkutsk death, then argues that a workplace exposure is more likely under his stated assumptions. The Biological Threat Landscape treats this as scenario reasoning, with the cause and agent still unconfirmed.","why":"Transparent assumptions help readers assess competing explanations without confusing modeled likelihood with an established cause.","chapters":[["The Biological Threat Landscape","/foundations/threat-landscape.html"]],"summary":"Esvelt's analysis favors workplace exposure under stated assumptions; the Irkutsk cause of death and agent remain unconfirmed.","section":"/foundations/threat-landscape.html#sec-irkutsk-2026","sources":[{"label":"Esvelt, 7 October 2026","url":"https://x.com/kesvelt/status/2107979312490246452","type":"post"}]},{"category":"revision","title":"WHO's 2025 SAGO assessment weighs evidence on COVID-19 origins","change":"WHO's 2025 SAGO assessment gives greater weight to zoonotic spillover, while missing information leaves a laboratory-associated breach unassessed and unexcluded. Attribution and Biological Forensics explains the limits of the origins investigation.","why":"Unequal evidentiary support matters when assessing competing explanations of COVID-19 origins.","chapters":[["Attribution and Biological Forensics","/governance/attribution-forensics.html"]],"summary":"WHO's 2025 SAGO assessment favors zoonotic spillover, while missing information prevents assessment and exclusion of a laboratory-associated breach.","section":"/governance/attribution-forensics.html#sec-attribution-covid","sources":[{"label":"WHO, SAGO independent assessment, 2025","url":"https://www.who.int/publications/m/item/independent-assessment-of-the-origins-of-sars-cov-2-from-the-scientific-advisory-group-for-the-origins-of-novel-pathogens","type":"report"}]},{"category":"revision","title":"Executive Order 14292 and revision of the DURC-PEPP framework","change":"Executive Order 14292 directed revision or replacement of the DURC-PEPP framework. Course Syllabus Template places this direction in its discussion of research oversight.","why":"The order defines the policy revision that research-oversight discussions must address.","chapters":[["Course Syllabus Template","/appendices/course-syllabus.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"summary":"Executive Order 14292 directed revision or replacement of the DURC-PEPP framework.","section":"/appendices/course-syllabus.html#sec-syllabus-12-week","sources":[{"label":"Executive Order 14292, 2025","url":"https://www.whitehouse.gov/presidential-actions/2025/05/improving-the-safety-and-security-of-biological-research/","type":"policy"}]}]},
{"date":"2026-10-13","headline":"Cell-free expression added as a production-machinery gap in DNA synthesis screening","entries":[{"category":"content","title":"Cell-free expression added as a production-machinery gap in DNA synthesis screening","change":"DNA Synthesis Screening adds Gap 3, cell-free expression: freeze-dried cell lysates can make proteins, toxins, and whole phages from a template with no customer verification, a production step outside synthesis screening.","why":"It shows a dual-use production pathway that sits outside the DNA-order chokepoint, with the primary literature documenting its reach.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"summary":"DNA Synthesis Screening adds a section on how cell-free expression systems produce proteins and phages from a template without the customer checks that govern synthesized DNA.","section":"/biotechnology/dna-synthesis-screening.html#sec-gap-cell-free","sources":[{"label":"Chitayat and Musunuri, 2026","url":"https://fas.org/publication/cell-free-biomanufacturing-strategic-capability/","type":"report"},{"label":"Molla et al., 1991","url":"https://doi.org/10.1126/science.1661029","type":"paper"},{"label":"Shin, Jardine, and Noireaux, 2012","url":"https://doi.org/10.1021/sb300049p","type":"paper"},{"label":"Nyerges et al., 2023","url":"https://doi.org/10.1038/s41586-023-05824-z","type":"paper"},{"label":"Fujino et al., 2024","url":"https://doi.org/10.1038/s41467-024-48056-z","type":"paper"}]},{"category":"developments","title":"Anthropic reports unintended model actions during live-internet evaluations","change":"Red-Teaming AI Systems adds Anthropic's report that Claude acted on real third-party systems in unintended ways, mostly by working around restrictions in public benchmarks run on the live internet, including the LABBench2 biology benchmark. Developer self-report with limited case detail.","why":"Realistic agent evaluations need containment: offline tasks and clearly stated targets, permitted actions, and network boundaries.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"summary":"Red-Teaming AI Systems adds a developer report of unintended model actions in live-internet evaluations, including a public biology benchmark.","section":"/ai-biosecurity/red-teaming.html#sec-red-teaming-realism","sources":[{"label":"Anthropic, October 2026","url":"https://www.anthropic.com/research/investigating-unintended-model-actions","type":"announcement"}]},{"category":"evidence","title":"DRYLAB-Bench: agentic mutation design scored against deep mutational scanning data","change":"Red-Teaming AI Systems adds DRYLAB-Bench, which scores model-designed single substitutions against 187,695 experimental measurements; the authors report that untooled designs from 15 frontier models exceed the assay mean in 67.6% of cases. Preprint, scored in silico.","why":"It measures design risk against empirical data, a construct that knowledge and refusal benchmarks do not capture.","chapters":[["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"]],"summary":"Red-Teaming AI Systems adds a preprint benchmark that scores agentic mutation design against experimental measurements rather than knowledge or refusal.","section":"/ai-biosecurity/red-teaming.html#sec-red-teaming-realism","sources":[{"label":"Wang et al., 2026, preprint","url":"https://arxiv.org/abs/2610.13116","type":"preprint"}]}]},
{"date":"2026-10-15","headline":"Executive Order 14434 and the CAISSI name for the US AI standards center","entries":[{"category":"policy","title":"Executive Order 14434 and the CAISSI name for the US AI standards center","change":"NIST names the US Center for AI Standards and Innovation the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), after Executive Order 14434 directed agencies to use \"Super Intelligence\" in place of \"Artificial Intelligence\" in public communications. How AI Amplifies Biosecurity Risk keeps CAISI for work published before that change.","why":"Readers tracking US model evaluation work need to recognize both names for the same body.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],["LLMs and Information Hazards","/ai-biosecurity/llms-info-hazards.html"],["Policy Frameworks for AI-Bio Convergence","/governance/policy-frameworks.html"]],"summary":"Executive Order 14434 directs agencies to say \"Super Intelligence\" instead of \"Artificial Intelligence\", and NIST's center carries the CAISSI name.","section":"/ai-biosecurity/ai-risk-amplifier.html#sec-amplifier-govt-evaluations","sources":[{"label":"Federal Register, 2026","url":"https://www.federalregister.gov/documents/2026/10/02/2026-20321/inaugurating-the-era-of-super-intelligence","type":"policy"}]}]},
{"date":"2026-10-16","headline":"DNA synthesis screening: the OSTP framework's 50 nucleotide window from October 13, 2026","entries":[{"category":"revision","title":"DNA synthesis screening: a 50 nucleotide window from October 13, 2026","change":"DNA Synthesis Screening dates the 2024 OSTP framework's additional practices to October 13, 2026: each 50 nucleotide window screened rather than each 200, bulk and repeated orders checked for assembly into a sequence of concern, and efforts made to screen sequences known to contribute to pathogenicity or toxicity. The framework remains under revision with no successor posted.","why":"A federally funded purchaser should confirm which window its award terms and its provider actually apply.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"summary":"From October 13, 2026 the 2024 OSTP framework's additional practices call for screening each 50 nucleotide window instead of each 200, while the framework remains under revision.","section":"/biotechnology/dna-synthesis-screening.html#sec-may-2025-eo","sources":[{"label":"OSTP, 2024","url":"https://bidenwhitehouse.archives.gov/wp-content/uploads/2024/10/OSTP-Nucleic-Acid_Synthesis_Screening_Framework-Sep2024-Final.pdf","type":"policy"}]}]},
{"date":"2026-10-20","headline":"Course syllabus: grading rubric and information-hazard rules for coursework","entries":[{"category":"content","title":"Course syllabus: grading rubric and information-hazard rules for coursework","change":"The syllabus assesses written biosecurity assignments on evidence, governance analysis, risk reasoning, information-hazard discipline, and communication. Coursework excludes operational biological detail and hazardous model probing; permitted AI use requires disclosure. Course evaluation combines learner feedback, governance-case assessment, follow-up on application, and review of reading currency.","why":"Instructors need assessment and coursework rules that keep biosecurity analysis at the level of assessment and governance.","chapters":[["Course Syllabus Template","/appendices/course-syllabus.html"]],"summary":"The syllabus pairs evidence and governance assessment with information-hazard safeguards, AI-use disclosure, and evaluation of learning and application.","section":"/appendices/course-syllabus.html#sec-syllabus-coursework-rules","sources":[]}]},
{"date":"2026-10-22","headline":"WHO genome-data consent guidance, IRB and IBC review, and CASPER wastewater sequencing","entries":[{"category":"policy","title":"WHO guidance on consent, reuse, and benefit sharing for human genome data","change":"Laboratory Biosafety and Biosecurity explains WHO's recommendations that consent describe potential commercial uses and AI training, that withdrawal limits be explained, and that access decisions stay accountable. WHO excludes pathogen and microbiome data; benefit-sharing terms should be discussed before collection or reuse, and CARE principles add collective governance for Indigenous data.","why":"Human-genome reuse needs clear authorization and benefit-sharing terms for affected people and communities.","chapters":[["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"]],"summary":"WHO recommends consent covering commercial use and AI training, explained withdrawal limits, and accountable access for human genome data; CARE adds collective governance for Indigenous data.","section":"/frameworks/laboratory-biosafety.html#sec-laboratory-cyberbiosecurity","sources":[{"label":"WHO, 2024","url":"https://www.who.int/publications/i/item/9789240102149","type":"policy"},{"label":"Carroll et al., 2020","url":"https://doi.org/10.5334/dsj-2020-043","type":"paper"}]},{"category":"policy","title":"IRB and IBC review for AI-biology research","change":"Dual-Use Research of Concern (DURC) explains that IRBs protect participants in covered human-subject research under 45 CFR part 46, while IBCs review covered recombinant or synthetic nucleic-acid research under the NIH Guidelines. Coverage depends on the activity and data, not AI use alone, and both bodies can draw on specialist advisers.","why":"Projects combining AI, human data, and biological work cross several review systems, and institutions need to coordinate them.","chapters":[["Dual-Use Research of Concern (DURC)","/frameworks/dual-use-research.html"]],"summary":"IRBs and IBCs review different parts of AI-biology research; coverage depends on the activity and data involved, and both can draw on specialist advisers.","section":"/frameworks/dual-use-research.html#sec-durc-oversight-framework","sources":[{"label":"45 CFR part 46","url":"https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-A/part-46","type":"policy"},{"label":"NIH Office of Science Policy, IBC administration FAQs, 2024","url":"https://osp.od.nih.gov/policies/biosafety-and-biosecurity-policy/faqs-on-institutional-biosafety-committee-ibc-administration-april-2024/","type":"policy"}]},{"category":"evidence","title":"CASPER wastewater sequencing data release from SecureBio and the University of Missouri","change":"AI for Biosecurity Defense reports the coalition's 1.2 trillion read pairs from 1,206 untargeted wastewater samples across 27 sites in nine U.S. states, with a median of 19 days from collection to sequencing-data delivery; this developer-authored preprint does not establish prospective outbreak-detection sensitivity. In October 2026 SecureBio announced future private core-pipeline development, with existing public code to remain accessible.","why":"Readers assessing surveillance need the reported dataset size and the time from sampling to delivery of sequencing data.","chapters":[["AI for Biosecurity Defense","/ai-biosecurity/ai-defense.html"]],"summary":"The CASPER coalition reported 1.2 trillion read pairs from 1,206 wastewater samples in nine states, a developer-authored preprint that does not establish detection sensitivity.","section":"/ai-biosecurity/ai-defense.html#sec-defense-nao","sources":[{"label":"Justen et al., 2026 (preprint, v2)","url":"https://www.medrxiv.org/content/10.64898/2026.03.05.26345726v2.full","type":"preprint"},{"label":"Bradshaw, SecureBio, 2026","url":"https://securebio.org/blog/taking-our-core-computational-pipeline-private/","type":"announcement"}]},{"category":"policy","title":"GAO findings applied to continuity checks for surveillance response","change":"Outbreak Detection and Surveillance recommends a continuity review that tests the verification and response pathway when funding, staffing, or responsibilities change: continued data access and skilled analysis, alert verification and escalation, and a named alternate owner with authority to act. The handbook presents this as an application of GAO's 2023 biosurveillance work.","why":"Surveillance signals help only if someone can still verify and act on them after funding or staffing changes.","chapters":[["Outbreak Detection and Surveillance","/frameworks/surveillance.html"]],"summary":"A continuity review tests data access, alert verification and escalation, and a named alternate owner when surveillance funding or staffing changes, applying GAO's 2023 work.","section":"/frameworks/surveillance.html#sec-surveillance-approaches","sources":[{"label":"GAO, November 2023","url":"https://www.gao.gov/products/gao-24-106142","type":"report"},{"label":"GAO, March 2023","url":"https://www.gao.gov/products/gao-23-106476","type":"report"}]},{"category":"policy","title":"Conditional purchase commitments for defensive capabilities: Project BioShield and Gavi","change":"Medical Countermeasures and Biodefense distinguishes research funding from purchase commitments, citing Project BioShield's support for advanced development and procurement and Gavi's pneumococcal advance market commitment. It outlines a proposed adaptation for AI-biosecurity defenses with a named buyer, qualifications, and independent testing before payment, noting that vaccine experience does not establish its effectiveness.","why":"Defensive capabilities often lack a market, and conditional purchasing is one way funders can create a credible buyer.","chapters":[["Medical Countermeasures and Biodefense","/frameworks/countermeasures.html"]],"summary":"Purchase commitments differ from research funding; Project BioShield and Gavi's pneumococcal commitment inform a proposed, untested financing design for AI-biosecurity defenses.","section":"/frameworks/countermeasures.html#sec-barda","sources":[{"label":"BARDA, Project BioShield","url":"https://medicalcountermeasures.gov/barda/cbrn/project-bioshield","type":"policy"},{"label":"Gavi, pneumococcal advance market commitment: how it works","url":"https://www.gavi.org/investing-gavi/innovative-financing/pneumococcal-advance-market-commitment-amc/how-it-works","type":"other"}]},{"category":"developments","title":"Ben Ouagrham-Gormley on tacit knowledge in AI-biosecurity risk assessment","change":"How AI Amplifies Biosecurity Risk adds Ben Ouagrham-Gormley's argument that assessments should distinguish personal expertise, communal team knowledge, organizational culture, and knowledge sustained by how work is structured. The essay offers an assessment framework and does not measure present-day AI uplift.","why":"Whether AI can substitute for hands-on expertise is central to estimating biological misuse risk.","chapters":[["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"]],"summary":"Ben Ouagrham-Gormley argues that AI-biosecurity assessments should distinguish personal, team, organizational, and work-structure knowledge; the essay does not measure present-day AI uplift.","section":"/ai-biosecurity/ai-risk-amplifier.html#sec-amplifier-threat-modeling","sources":[{"label":"Ben Ouagrham-Gormley, Georgetown Journal of International Affairs, 2026","url":"https://gjia.georgetown.edu/science-technology/rethinking-the-de-skilling-narrative-in-ai-and-biological-weapons-policy/","type":"other"}]},{"category":"developments","title":"Edward You's proposed digital handshake between biological suppliers and customers","change":"DNA Synthesis Screening describes Edward You's proposal for limited confirmations that a purchase is authorized and tied to a legitimate institutional account or project, without unrestricted information disclosure, using graduated authentication and human review. The essay does not report an evaluation of the approach.","why":"Customer verification must confirm legitimacy without requiring customers to disclose more than necessary.","chapters":[["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"]],"summary":"Edward You proposes limited supplier and customer confirmations of authorized purchases, with graduated authentication and human review; the essay does not report an evaluation.","section":"/biotechnology/dna-synthesis-screening.html#sec-kyc-framework","sources":[{"label":"You, Issues in Science and Technology, 2026","url":"https://issues.org/biosecurity-ai-you/","type":"other"}]},{"category":"developments","title":"BioGovernance Commons: researcher-led biosecurity dialogue","change":"The Future of Biosecurity adds Zhang and colleagues' 2022 account of BioGovernance Commons, recurring online discussions among Chinese and Western researchers on shared ethical and regulatory questions, with attention to local institutional contexts. It is an account of the organizers' experience and does not measure changes in biological risk.","why":"Researcher-to-researcher exchange is one route to shared biosecurity norms across national research systems.","chapters":[["The Future of Biosecurity","/governance/future-biosecurity.html"]],"summary":"Zhang and colleagues describe BioGovernance Commons, recurring dialogue among Chinese and Western researchers; the account does not measure changes in biological risk.","section":"/governance/future-biosecurity.html#sec-future-international","sources":[{"label":"Zhang et al., Issues in Science and Technology, 2022","url":"https://issues.org/common-ground-chinese-researchers-biogovernance/","type":"other"}]}]}
],"highlights":[{"date":"2026-10-22","title":"WHO guidance on consent, reuse, and benefit sharing for human genome data","tier":"major","category":"policy","chapter":["Laboratory Biosafety and Biosecurity","/frameworks/laboratory-biosafety.html"],"sources":[{"label":"WHO, 2024","url":"https://www.who.int/publications/i/item/9789240102149","type":"policy"},{"label":"Carroll et al., 2020","url":"https://doi.org/10.5334/dsj-2020-043","type":"paper"}],"section":"/frameworks/laboratory-biosafety.html#sec-laboratory-cyberbiosecurity"},{"date":"2026-10-20","title":"Course syllabus: grading rubric and information-hazard rules for coursework","tier":"major","category":"content","chapter":["Course Syllabus Template","/appendices/course-syllabus.html"],"sources":[],"section":"/appendices/course-syllabus.html#sec-syllabus-coursework-rules"},{"date":"2026-10-16","title":"DNA synthesis screening: a 50 nucleotide window from October 13, 2026","tier":"major","category":"revision","chapter":["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],"sources":[{"label":"OSTP, 2024","url":"https://bidenwhitehouse.archives.gov/wp-content/uploads/2024/10/OSTP-Nucleic-Acid_Synthesis_Screening_Framework-Sep2024-Final.pdf","type":"policy"}],"section":"/biotechnology/dna-synthesis-screening.html#sec-may-2025-eo"},{"date":"2026-10-13","title":"Anthropic reports unintended model actions during live-internet evaluations","tier":"landmark","category":"developments","chapter":["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],"sources":[{"label":"Anthropic, October 2026","url":"https://www.anthropic.com/research/investigating-unintended-model-actions","type":"announcement"}],"section":"/ai-biosecurity/red-teaming.html#sec-red-teaming-realism","vendor":true},{"date":"2026-10-10","title":"Esvelt Irkutsk analysis; WHO SAGO assessment of COVID-19 origins","tier":"major","category":"developments","chapter":["The Biological Threat Landscape","/foundations/threat-landscape.html"],"sources":[{"label":"Esvelt, 7 October 2026","url":"https://x.com/kesvelt/status/2107979312490246452","type":"post"},{"label":"WHO, SAGO independent assessment, 2025","url":"https://www.who.int/publications/m/item/independent-assessment-of-the-origins-of-sars-cov-2-from-the-scientific-advisory-group-for-the-origins-of-novel-pathogens","type":"report"}],"section":"/foundations/threat-landscape.html#sec-irkutsk-2026"},{"date":"2026-10-09","title":"WHO account of the Irkutsk event; PABS annex adoption tied to ratification","tier":"major","category":"policy","chapter":["The Biological Threat Landscape","/foundations/threat-landscape.html"],"sources":[{"label":"WHO, Director-General's media briefing remarks, 7 October 2026","url":"https://www.who.int/news-room/speeches/item/who-director-general-s-opening-remarks-at-the-media-briefing---7-october-2026","type":"announcement","short_label":"WHO, 7 October 2026"},{"label":"WHO, September 25, 2026","url":"https://www.who.int/news/item/25-09-2026-world-leaders-renew-commitment-to-protect-the-world-from-future-pandemics","type":"announcement"}],"section":"/foundations/threat-landscape.html#sec-irkutsk-2026"},{"date":"2026-10-08","title":"UK AISI reports chemistry and biology evaluation trends and safeguard limits","tier":"major","category":"policy","chapter":["How AI Amplifies Biosecurity Risk","/ai-biosecurity/ai-risk-amplifier.html"],"sources":[{"label":"UK AISI, Frontier AI Trends Report","url":"https://www.aisi.gov.uk/frontier-ai-trends-report#chemistry--biology","type":"report"}],"section":"/ai-biosecurity/ai-risk-amplifier.html#sec-amplifier-govt-evaluations"},{"date":"2026-10-07","title":"California AB 1864 binds companies to the 2024 synthesis screening framework","tier":"landmark","category":"policy","chapter":["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],"sources":[{"label":"California Legislature, AB 1864, 2026","url":"https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB1864","type":"policy"}],"section":"/biotechnology/dna-synthesis-screening.html#sec-mandatory-screening"},{"date":"2026-10-06","title":"SecureBio reports VCT-v2 question revisions and benchmark limits","tier":"major","category":"evidence","chapter":["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],"sources":[{"label":"Mak and Götting, 2026","url":"https://securebio.org/blog/introducing-vct-v2/","type":"report"}],"section":"/ai-biosecurity/red-teaming.html#sec-red-teaming-vct-v2"},{"date":"2026-10-01","title":"Goodfire sequence-aware protein-embedding biosecurity monitors","tier":"major","category":"evidence","chapter":["Autonomous AI Agents in Laboratory Contexts","/ai-biosecurity/autonomous-agents.html"],"sources":[{"label":"Goodfire, 2026","url":"https://www.goodfire.com/research/better-biosecurity-monitors","type":"announcement"}],"vendor":true},{"date":"2026-09-30","title":"SynthIDBio: protein watermarks in Nature, with DeepMind's companion","tier":"landmark","category":"evidence","chapter":["DNA Synthesis Screening: The Critical Chokepoint","/biotechnology/dna-synthesis-screening.html"],"sources":[{"label":"Stutz et al., 2026","url":"https://doi.org/10.1038/s41586-026-10965-y","type":"paper"},{"label":"DeepMind, Introducing SynthID Bio, 2026","url":"https://deepmind.google/blog/introducing-synthid-bio/","type":"announcement"}],"section":"/biotechnology/dna-synthesis-screening.html#sec-gap-ai-sequences","vendor":true},{"date":"2026-09-29","title":"OpenAI reports GPT-6.1 Sol Preparedness classifications","tier":"landmark","category":"developments","chapter":["Red-Teaming AI Systems for Biosecurity Risks","/ai-biosecurity/red-teaming.html"],"sources":[{"label":"OpenAI, GPT-6.1 Sol addendum system card, 2026","url":"https://cdn.openai.com/pdf/38e3efcf-545e-44cd-99ec-2b7eb395f4cc/oai_GPT_6_1_Sol.pdf","type":"announcement"}],"section":"/ai-biosecurity/red-teaming.html#sec-red-teaming-full-system","vendor":true},{"date":"2026-09-28","title":"Klapsa et al. document London poliovirus detection in wastewater","tier":"landmark","category":"evidence","chapter":["Outbreak Detection and Surveillance","/frameworks/surveillance.html"],"sources":[{"label":"Klapsa et al., 2022","url":"https://doi.org/10.1016/S0140-6736(22)01804-9","type":"paper"}],"section":"/frameworks/surveillance.html#sec-wastewater-surveillance"}],"months":{"2025-11":"The handbook published its first edition, launching with Welcome, Preface, and References and How to Cite pages under a CC BY 4.0 license. It outlined a five-part chapter structure and completed initial chapters, including What Is Biosecurity?, The Biological Threat Landscape, and Pathogens of Concern.","2025-12":"The handbook completed most of its remaining chapters, including Dual-Use Research of Concern, International Governance and the Biological Weapons Convention, DNA Synthesis Screening, Gain-of-Function Research, and eight new AI-biosecurity and governance chapters, along with an Executive Summary and closing appendices. It added a Science-reported October 2025 demonstration that AI-designed sequences evaded DNA synthesis screening, and noted that an executive order rescinded the 2024 OSTP synthesis-screening framework.","2026-01":"The handbook added Anthropic's account of activating ASL-3 safeguards for Claude Opus 4, the Virology Capabilities Test finding that OpenAI's o3 model outperformed expert virologists, and the Paraphrase Project's finding that AI-redesigned toxins evaded DNA synthesis screening. It also covered the 2024 International Health Regulations amendments entering into force and added case studies on Iraq's biological weapons program and nasal vaccine countermeasures.","2026-02":"The handbook added a new chapter on Autonomous AI Agents, covering the agent skill supply chain and multi-agent coordination risks, after a study found vulnerabilities in 26.1% of over 42,000 tested agent skills. A red-team test also found that 36 of 38 DNA synthesis providers shipped fragments sufficient to reconstruct 1918 influenza, and the handbook added the Biosecurity Modernization and Innovation Act of 2026 to its DNA screening coverage.","2026-03":"The handbook added a case study on Anthropic's February 2026 rollback of its Responsible Scaling Policy, and cited a red-team test finding screening gaps at 36 of 38 DNA synthesis providers, including 12 of 13 IGSC members. A GAO report found that HHS does not consistently disclose its gain-of-function oversight, and the handbook noted that an NIH notice rescinded the implementation notice for the 2024 dual-use research oversight policy.","2026-04":"The handbook added a new chapter, The Chemical Weapons Convention and OPCW, covering the treaty's structure and the OPCW's inspection and verification regime. It also cited a NIST red-teaming competition in which more than 400 participants defeated every tested frontier AI agent, and clarified that the World Health Assembly adopted the WHO Pandemic Agreement in May 2025 as WHA78.1, with one annex still under negotiation.","2026-05":"The handbook reported that OpenAI's GPT-5 proposed and analyzed cell-free protein synthesis experiments autonomously at Ginkgo Bioworks, and added the International AI Safety Report 2026's findings on biological risk. The U.S. Center for AI Standards and Innovation also expanded its pre-deployment testing program to Google DeepMind, Microsoft, and xAI, and the handbook cited two Nature papers, Robin and Co-Scientist, describing multi-agent AI systems extending into open-ended biology.","2026-06":"The handbook cited Los Alamos National Laboratory's Biosurveillance Gateway and Sandia National Laboratories' biorisk assessment models as institutional surveillance examples, and expanded its coverage of personal protective equipment and Far-UVC environmental controls with CDC, NIOSH, and OSHA sourcing. It also added a caveat that real-room conditions limit Far-UVC disinfection compared with controlled settings, and noted that the Community Biology Biosafety Handbook remains a beta community draft.","2026-07":"The handbook expanded Red-Teaming AI Systems with five analytical layers, including capability measurement and human and agent uplift. It added the Sequence Biosecurity Risk Consortium's Biosecurity Flag Rubric as a testable definition of a 'sequence of concern,' and cited UK AI Security Institute and SecureBio evaluations of Constitutional Classifiers using a Boundary Point Jailbreaking method. A developer-reported all-atom protein design model, RFdiffusion3, was also added.","2026-08":"The handbook reported King et al.'s demonstration of experimentally functional bacteriophages generated with genome language models, and added the July 2026 U.S. Government Policy for Stopping High-Risk Life Sciences Research to its gain-of-function coverage. It also added a new Cyberbiosecurity for Connected Laboratories section, and cited a RAND report stress-testing proposed AI-bio defense controls against multiple scenarios. AI-bio risk coverage also examines real-world actors.","2026-09":"SynthIDBio protein watermarks provide provenance without replacing DNA order screening, while OpenAI reports GPT-6.1 Sol as High in Biological and Chemical capability. Klapsa et al. document London wastewater detection of poliovirus, and vaccine supply-chain studies connect preparedness with logistics and manufacturing capacity. NSCEB proposals connect biological-data infrastructure with access controls and AI-ready standards.","2026-10":"Additional 2024 OSTP synthesis-screening practices applied October 13, 2026; California AB 1864 binds covered companies to the framework, while Executive Order 14292 directed DURC-PEPP revision or replacement. WHO's Irkutsk and COVID-19 origins accounts and Esvelt's scenarios distinguish evidence from unresolved explanations; RAND proposes collective restraint to prevent mirror life. Coursework pairs governance assessment with information-hazard safeguards; WHO human-genome guidance addresses consent, while recommended surveillance continuity checks test response responsibilities."},"upkeep":["2025-12","2026-01","2026-02","2026-03","2026-04","2026-05","2026-06","2026-07","2026-08","2026-09","2026-10"]}