AI safety × biosecurity

In “We Must Pace the Frontier,” Amodei proposes permanent external evaluators with ongoing access inside labs, followed by industry and government coordination on development pace. The essay states an Anthropic commitment; it is not evidence that the arrangement is already operating.Questions to follow: Who selects and pays evaluators, what can they inspect and publish, and what happens when findings call for delay?

Fathom proposes a marketplace of independent verification organizations, with government-set goals and specialist assessment. This differs from embedding reviewers inside a lab: the debate also concerns certification, incentives, and who has authority to act on findings.Questions to follow: How would assessor independence, common standards, liability, and public accountability work?

METR’s May 2026 Frontier Risk Report examines agent capabilities and misalignment risks; Apollo’s Evals Gap examines limits of evaluation evidence. Read measured results alongside assumptions about access, elicitation, and how tests transfer to deployment.Questions to follow: What was actually tested, what was inaccessible, and what uncertainty remains after a good result?

Amodei’s “The Adolescence of Technology” (January 2026) presents an industry leader’s case for managing powerful AI risks. Compare its biological-risk framing with the National Academies’ synthesis and RAND’s layered biosecurity strategy, rather than treating a forecast as settled evidence.Questions to follow: Which capability claims translate into real-world risk, and where should model safeguards, screening, and resilience each contribute?

Their essay “Do AI Risks Require Extraordinary Government Intervention?” argues for stronger societal defenses and a higher bar for exceptional restrictions. It challenges parts of the frontier-risk policy framing while accepting misuse risks, including biological risks.Questions to follow: When are restrictions justified, and which defenses remain effective as capabilities spread?

NTIA’s report maps benefits and marginal risks of widely available weights, including research access, competition, and misuse. It offers a structured entry point into a continuing debate; its 2024 recommendations are not a statement of current law or current model capabilities.Questions to follow: What risks are added by releasing weights, which benefits require openness, and what evidence would change the policy choice?

“Artificial Power” frames AI governance around concentrated corporate power and public agency. This perspective widens the safety discussion to institutional accountability and existing societal harms, alongside catastrophic-risk concerns.Questions to follow: Whose harms count in safety decisions, and who has the power to challenge deployment?

Crystal Grant reviews 2025 developments in AI and biology and offers projections for 2026, discussing capabilities, biological risk, and governance. A dated policy perspective; projections and interpretations are distinct from established findings.

GPAI systemic-risk duties and AI-system rules. Use this Commission page for the amended 2026 implementation timeline.

Proposed biotechnology framework including biosecurity and AI-enabled innovation. The Council agreed a mandate on the directive strand in June 2026; that does not enact the broader regulation proposal.

Licensing and oversight of controlled pathogen and toxin activities; implementation page distinguishes changes coming later.

National framework spanning biotechnology, laboratories, biological resources, and biological threats.

AI governance and trust framework; ministry notice covers 2026 amendments and proposed implementing-decree changes.

Government overview of the research, development, adoption, and risk-governance framework. Japanese text.

AI regulatory proposal and Senate legislative history. Follow onward legislative action before treating any provision as law.

National security controls for specified biological agents; relevant to the physical side of AI×Bio.

Directs changes to oversight and federal funding of high-risk biological research; not a statute.

Sections 241–248 cover biomanufacturing, biological data for AI and responsible biotechnology; §§6611–6616 address intelligence capabilities and biological-data security.

Restricts federal procurement and funding involving biotechnology companies of concern. A supply-chain/national-security law; its operative dates depend on implementing steps.

Restricts specified transactions exposing bulk sensitive personal data—including genomic and other covered biological data—to countries of concern and covered persons.

Controls selected biotechnology equipment and related technology, explicitly addressing biological data generation and AI-enabled national-security risks.

Baseline US oversight of possession, use and transfer of specified biological agents and toxins. Connects AI×Bio governance to existing laboratory regulation.

Proposes White House biotechnology coordination and a national strategy; a central NSCEB proposal for organizing federal bioeconomy policy.

Would task NIST with biological-data standards and frameworks for AI use. A proposed mandate to create standards, not a published technical standard.

Proposes an NSF cloud-laboratory network for automated experimentation and biological data generation; relevant to governing AI-connected laboratory infrastructure.

Would establish a DOE entry point for AI-ready biological datasets with security and access safeguards; complements the proposed NIST standards and cloud-lab network.

Federal synthesis-security proposals. Senate text includes screening regulation, standards and governance testing; House text is distinct and should be compared separately.

Two NSCEB-backed bills would expand CISA protection of biotechnology/biomanufacturing infrastructure and sensitive biological data. Useful watchlist for cyberbiosecurity.

Transparency in Frontier Artificial Intelligence Act: developer safety frameworks, incident reporting, and whistleblower protections.

RAISE Act frontier-model transparency and safety requirements. Read with the enacted S8828 / A9449 chapter amendment, which replaces provisions and changes implementation details.

Senate amendment proposes frontier-AI transparency and assurance requirements. Latest displayed bill history shows conference proceedings, not enactment.

Proposes frontier-developer risk disclosures and safety obligations. Primary introduced text; current action history needs rechecking.

Independent AI verification framework and an auditor registry with independence and transparency standards. A direct link between AI safety law and third-party assurance; implementation details require the statutory text.

Frontier AI risk frameworks, incident reporting and assurance obligations, including catastrophic biological risks. Core developer requirements have later commencement dates.

Replaces the earlier SB24-205 framework with requirements for automated decision-making in consequential decisions; broader assurance relevance.

Texas Responsible Artificial Intelligence Governance Act: prohibited uses, disclosures, and governance; not a frontier-biosecurity statute.

Earlier frontier-model safety proposal. Governor’s veto message; this bill did not become law.

Find additional enacted and pending state bills from 2025 onward, including responsible use, discrimination, and healthcare.

Would require covered gene-synthesis providers and benchtop-equipment manufacturers to follow specified screening safeguards and publish compliance attestations.

Would regulate gene-synthesis providers and equipment through state health rules, including customer and sequence screening and records.

Recent policy issued under EO 14292. Read alongside agency implementation; do not rely on 2024 DURC/PEPP materials alone.

Responsibilities for synthesis providers, users, and benchtop equipment.

Responsible life sciences and governance of dual-use research across institutions and governments.

Biosecurity across the material, information, and technology lifecycle; includes emerging technology risks.

Safety/security, transparency, and copyright measures supporting compliance with binding AI Act duties.

National biological-risk priorities and institutional responsibilities.

Regional AI governance, development, and capacity priorities; not a continent-wide AI law.

Track screening guidance, implementation resources and provider attestations. Read alongside EO14292 and current agency funding requirements.

Responsible-use principles covering biotechnology, AI convergence, biological data and defence; useful for allied governance and security cooperation.

Bioeconomy, sustainable biomanufacturing and innovation policy broadening the directory beyond US/EU security debates; relevant to Bio-AI hubs and infrastructure.

Government framework for engineering-biology research, commercialization, infrastructure, skills and responsible innovation; pairs industrial policy with biological security.

Core US laboratory biosafety reference for protocol-driven risk assessment and institutional practices. Advisory guidance rather than a standalone regulation.

Risk-based framework for safe laboratory work, with companion monographs. A foundational biosafety reference alongside WHO's separate biosecurity guidance.

Organizes AI risk work into Govern, Map, Measure, and Manage. Start here for a risk-management structure; not a certification scheme.

Adds generative-AI risks and suggested actions to the AI RMF. Use with the core framework for generative models and applications.

Suggested actions for applying the AI RMF’s four functions. Useful for translating a framework into day-to-day governance practices.

Secure-development practices for generative AI and dual-use foundation models. Use with SP 800-218; focuses on development security.

Track voluntary guidelines for advanced models, systems, and agents. Check each document’s version and draft/final status.

Containment, operations, and verification requirements for regulated facilities. Current edition and biosecurity addendum continue to apply pending alignment with 2026 law changes.

Governance guidance for AI developers, providers and business users, with checklists and worksheets. Japanese page includes an English tentative translation.

Six practices covering accountability, impacts, testing and monitoring. Updated successor to the 2024 Voluntary AI Safety Standard.

Governance guidance for generative and agentic AI, including human accountability, deployment controls and managing agent autonomy.

Basic security requirements for generative AI services. Chinese national-standard record; useful for comparing national assurance approaches.

An auditable validation framework for AI used in diagnosis, treatment and health-condition management; a sector-specific assurance example.

Requirements for an AI management system. Certification is not proof that a particular model or biological application is safe. Full text paid.

Requirements for bodies auditing and certifying AI management systems under ISO/IEC 42001. Full text paid.

AI-specific risk management and integration into organizational processes. Full text paid.

Biorisk management for laboratories and related organizations. ISO lists a 2024 amendment and a revision in development. Full text paid.

Production and quality-control requirements for synthesized gene fragments, genes, and genomes; scope differs from a comprehensive biosecurity regime. Full text paid.

Baseline cybersecurity requirements across the AI lifecycle. Complements AI risk management with security controls for models and systems.

Guidance for assessing and documenting AI impacts on people and society throughout the lifecycle; complements 42001 and 23894. Full text paid.

Track published AI standards and projects in development, rather than discovering each standard separately.

Accreditation of certification bodies for ISO/IEC 42001; explains the layer of oversight above certification.

Find European AI standards supporting risk management and conformity assessment. Check each document's stage and EU Official Journal citation separately.

A shared vocabulary for evaluation, audit, certification, and assurance mechanisms.

Examples of assurance in practice, spanning safety, robustness, transparency, and accountability.

Independent assurance ecosystem, professional skills, and institutional development.

Infrastructure for reproducible model evaluations. A tool, not a certification or a biosecurity verdict.

Benchmarking and red-teaming of AI systems and LLM applications; general assurance rather than specialized biological-risk certification.

Why current evaluation evidence can be insufficient for high-stakes safety decisions.

Scientific foundation for frontier governance, including transparency and evidence supporting safety claims.

Company thresholds and safeguards for frontier risks. A developer commitment, not independent assurance.

Capability thresholds, evaluations, and safeguards; distinguish company claims from external verification.

An early controlled comparison of LLM-assisted and unassisted planning. Useful for understanding uplift-study design and uncertainty; its findings concern the tested systems and do not establish the safety of current models.

Practices for transparent, reproducible automated evaluations of language models and agents; explains where benchmarks are unsuitable.

Framework for comparing biological AI tools' capabilities, accessibility and misuse relevance. Helps prioritize review before funding or release.

Framework for identifying, measuring and mitigating deliberate misuse of foundation models, including biological and chemical risk considerations.

Sequence and customer-screening resources for DNA/RNA providers.

Screening infrastructure and implementation documentation for synthesis providers.

Access-governance options for biological AI tools with different benefits and risks.

Core prohibition on biological weapons; UNODA overview of the treaty and its institutions.

Domestic controls to prevent non-state actors acquiring nuclear, chemical, or biological weapons; committee fact sheet.

Coordination of controls on chemical/biological materials, technology, and equipment; domestic rules implement controls.

Human-rights, democracy, and rule-of-law framework. Check ratification and applicability country by country.

Cross-sector working groups and cooperation, including biological evaluation practice.

Route into pandemic prevention, access and benefit sharing, and governance of pathogen-related data. Adoption is distinct from entry into force.

Shared sequence- and customer-screening practices for synthesis providers and equipment manufacturers. Voluntary industry protocol, distinct from government rules.

Practical recommendations for measurable synthesis screening, implementation support and conformity assessment, developed through cross-sector workshops.

Track NIST screening measurement work, standards-related projects and test resources; connects synthesis safeguards with national measurement science.

Structured prompts to identify potential dual-use concerns in life-science research and start institutional review; not a regulatory clearance.

Structured review of life-science proposals before funding. Its stated scope excludes AI-enabled biological design, synthesis and optimization proposals.

Principles for responsible life-science research, institutions, training and publication; useful when developing research codes of conduct.

AI×Bio risk evaluation, AI security, bioresilience, and technology governance; builds on RAND’s Meselson and TASP centers.

Governance initiatives at the intersection of AI and the life sciences.

AI×Bio governance and public-health preparedness expertise.

Research on biological risk, threat modeling, and governance of advanced AI.

AI and biosecurity policy analysis, including DNA synthesis governance.

AI and biosecurity policy, including analysis of proposed EU biotechnology safeguards.

Studies biological and other systemic security risks and runs a Global Biosecurity Accelerator. Useful for biodefense policy, international resilience, and the relationships among emerging technologies and strategic threats.

Publishes assessments and recommendations on US biodefense, including the National Blueprint for Biodefense and AI-related priorities. A major route to policy proposals and accountability questions across the biosecurity system.

Publishes bioeconomy and biosecurity policy analysis, including commentary on NSCEB recommendations and proposals for US biological infrastructure. Useful for understanding policy options and implementation gaps.

Conducts AI safety research, risk monitoring, policy engagement, and international dialogue from Beijing and Singapore. Its country reports and biological-risk work help readers follow Chinese and regional governance perspectives.

International Biosecurity and Biosafety Initiative for Science; synthesis screening and responsible science.

Connects academic, government, and industry engineering-biology communities on security, policy, and technical standards. Its security program covers synthesis-screening assurance, responsible research, and training.

Coordinates synthesis providers around customer and sequence screening and shared biosecurity practices. A key industry implementation actor alongside independent screening tools and government requirements.

AI biological risk evaluation, safeguards, and pandemic prevention.

Frontier model capabilities and risks, particularly autonomous task performance.

Model evaluations, including their limitations and implications for safety decisions.

Researches and tests frontier AI safeguards, robustness, and risk mitigation. Its current work includes a consortium developing CBRN evaluations for the EU AI Office.

Works on societal-scale AI risks through technical research, training, and safety standards advocacy. Offers benchmarks, research infrastructure, and educational routes into AI safety.

Researches AI accountability, governance, and the institutions needed for trustworthy AI. Its work on assurance and professionalization complements frontier-risk evaluation with broader societal and deployment concerns.

Develops methods for evaluating and mitigating risks across the AI-system lifecycle in high-stakes settings. Useful for practical assurance planning, governance, and evidence collection.

Provides tools and methods for constructing maintainable assurance arguments linking claims to evidence and assumptions. Useful for moving from a checklist of risks to an explicit safety or trustworthiness case.

Frontier AI evaluations and safety research; formerly the AI Safety Institute.

AI testing, standards, and research including biological and chemical risks.

Open-source AI testing ecosystem, including traditional and generative AI.

Expert advice on dual-use governance and mitigation of biological risks.

Science, technology, and biological weapons governance research.

Supports national biosafety and biosecurity laws, institutional systems, training, and certification across African Union member states. Its current strategy explicitly addresses emerging technologies including AI and synthetic biology.

Develops AI safety evaluation methods and guidance and coordinates public, private, and international work. A route to Japanese evaluation guides and policy updates.

Supports research and practical tools for understanding and reducing advanced-AI risks. Use the government overview to distinguish its mandate from CIFAR’s affiliated research program.

Brings together French expertise in AI evaluation, security, and measurement. Provides an institutional entry point for France’s AI safety work and its links to international evaluation networks.

Congressional body linking biotechnology, national security, AI-ready biological data, biomanufacturing and governance; start with its report and implementation tracker.

Publishes policy proposals and AI safety assessments, funds research, and advocates measures to reduce catastrophic technology risks. Useful for tracking a prominent civil-society perspective, with its advocacy position clearly labeled.

Develops AI governance institutions and promotes independent verification organizations (IVOs). A route to proposals linking assurance, public trust, and innovation, including its work on California SB 813.

Develops policy and advocates safety standards to address catastrophic AI risks. Its model legislation, bill endorsements, and advocacy network help track concrete policy positions.

Advocates proactive AI governance balancing public protection and innovation. Publishes federal frontier-governance proposals and analysis of accountability and national-security policy.

Campaigns to pause dangerous AI development and for public oversight. A route to the precautionary and public-mobilization side of the debate; its positions are advocacy rather than a scientific consensus.

Examines corporate power, accountability, and societal impacts of AI. Brings labor, civil-rights, and political-economy perspectives to debates often framed primarily around frontier capabilities.

Combines technical research and policy analysis on national, international, and corporate governance of advanced AI. A route to research on institutions, oversight, and technically feasible governance mechanisms.

Studies biological threats, research governance, and the social and institutional conditions needed to reduce risk. Connects biology, engineering, medicine, law, and international-security expertise.

Develops theory, experiments, and policy relevant to beneficial and controllable AI. A general AI-safety research anchor, rather than a specialist AI–biosecurity institution.

Funds biosecurity and pandemic preparedness as well as work on navigating transformative AI. Its program strategies and grant database help identify both priorities and organizations working in the field. Renamed in November 2025; check individual calls for current application windows.

Researches, funds, and coordinates projects to prevent catastrophic pandemics, including safeguards and governance. Useful for understanding philanthropic priorities and implementation gaps. Does not accept unsolicited proposals.

Advises donors and supports work on advanced AI and biological risks. Grant reports offer a route to organizations and theories of change across the safety ecosystem. Its Emerging Challenges Fund is closing and is not accepting new donations.

Offers structured AI safety, AI governance, and biosecurity courses, alongside career-transition support and grants. A strong starting point for newcomers and professionals moving between these fields.

Trains and places emerging-technology experts in government and think tanks, including AI and biotechnology specialists. Its public career guides explain policy institutions and routes into public service. The 2027 fellowship application round is closed.

Connects synthetic-biology education with biosafety, biosecurity, risk management, and responsible innovation. Useful for finding practitioner communities and teaching resources at the AI–biology intersection.

Professional network for people implementing biosafety and biosecurity in laboratories and institutions. Provides training, conferences, and practitioner resources, including emerging AI and cyberbiosecurity topics.

Links regional and national biosafety associations and develops professional capacity in biorisk management. Useful for locating local partners and understanding the workforce needed to implement governance.

Convening and research organization focused on international cooperation to reduce advanced-AI risks. Its International Dialogues on AI Safety offer a route to cross-border safety discussions.

ERA describes this as a fully funded, in-person research fellowship in Cambridge run in partnership with the Cambridge Biosecurity Hub. The 10-week program supports fellows pursuing AI-and-biosecurity research, placements, or potential new organizations.

Broad safety evidence, including biological risk and risk management. Benchmark performance and real-world harm are distinct questions.

Early framing of the intersection and governance options; read newer evidence alongside it.

Bhuvana Sudarshan and Rebecca Hersman on accident prevention as a complement to misuse controls.

IBBIS people directory and areas of work.

Research-team directory; includes current and past members, which are labeled separately.

King’s College London profile and research background.

Contributors to NTI’s funder risk-assessment guidance. Its scope excludes AI-enabled biological design and synthesis proposals.

Maps the convergence of biological design, AI, and automation, with governance options for biosecurity, data supply chains, and human oversight. Useful for framing AI×Bio within the wider bioeconomy.

Compares national strategies and policy instruments for responsible biotechnology innovation. A route into bioeconomy governance beyond US security policy; recommendations, not binding rules.

Six-pillar policy map for US biotechnology competitiveness and security, including biological data, responsible innovation, workforce and allied cooperation.

Practical policy options for biotechnology market pathways and regulatory coordination across medical, agricultural and industrial uses; complements risk-focused governance.

Connects model safeguards, synthesis controls, monitoring and resilience into a layered strategy, with concise policymaker and industry infographics.

Evidence synthesis and recommendations on biological AI tools, sensitive biological data and defensive opportunities, focused on pandemic-scale threats.

Recurring assessments of capabilities, emerging developments, and governance questions.

Discover national AI initiatives across jurisdictions; verify legal status against primary legal records.

Reported incidents and hazards; reporting coverage is not the same as prevalence.

Publicly documented autonomy and deception incidents, with evidence limitations noted.

Follow evaluations and methods work from a public frontier-AI research institute.

Implementation discussions and meeting records for the General-Purpose AI Code.

Country profiles and comparisons of measures implementing the Biological Weapons Convention. Use to discover national biological-weapons laws, then confirm operative text with the jurisdiction.

Search a broader library of practical biosecurity guidance and training. Developed through Georgetown, WOAH, and Canadian support; a useful next stop when this selective directory is too narrow.

Follow which commission recommendations became bills, executive actions or enacted provisions; use the legislative drafts to spot proposals not yet introduced.