AIxBio Resources
00
Biosecurity Strategy
-
The Four Pillars: A Hypothesis for Countering Catastrophic Biological Risk
A strategic hypothesis: physical protection and early warning preserve response capacity while pathogen-specific medical countermeasures are developed.
Andrew Snyder-Beattie · Defenses in Depth · 1 October 2025 -
10 big projects for reducing bio x-risk
Ten project proposals with objectives and success criteria, including PPE stockpiles, independent equipment testing and assessment of biosecurity safeguards.
Chris Bakerlee · Defenses in Depth · 9 May 2026 -
Building a Defense-in-Depth Biosecurity Strategy for the AI Era
Nine complementary mitigations, principally before release: information and material access controls, deterrence and proactive detection of AI-enabled biological misuse.
RAND · 18 August 2026
Industry perspective
-
Biodefense in the Intelligence Age
An action plan for trusted access to biological AI, countermeasure research and early warning, with evaluation of defensive benefits and misuse risks.
OpenAI · 4 June 2026 -
Advanced AI Framework: biological resilience
Proposals linking developer oversight with synthesis screening, surveillance, respiratory protection and countermeasure capacity.
Anthropic · June 2026 · pp. 13–16 -
Our approach to bioresilience
A programme of model safeguards, metagenomic surveillance and AI-assisted countermeasure development, organised around prevention, detection and response.
Google DeepMind & Isomorphic Labs · 16 July 2026
01
Understanding biosecurity risk
-
Understanding AI-Facilitated Biological Weapon Development
An introduction to biological AI tools, their maturity and potential contribution to misuse, with an initial map of intervention points.
Centre for Long-Term Resilience · 2023 · Report -
The near-term impact of AI on biological misuse
A framework for specifying and testing claims about AI-enabled uplift, distinguishing model capabilities, access conditions and threat-actor characteristics.
Centre for Long-Term Resilience · 2024 · Report -
How I think about catastrophic biological risk (part II): risk breakdown by type of prevention
Distinguishes risk from civilian accidents, state programmes and non-state misuse, then examines how assumptions about their relative contributions inform priorities for synthesis screening, AI safeguards and other preventive measures.
Andrew Snyder-Beattie · Defenses in Depth · 15 May 2026 · Analysis
Further reading · 4
- The Age of AI in the Life Sciences: Benefits and Biosecurity Considerations National Academies of Sciences, Engineering, and Medicine · 2025 · Consensus report
- Capability-Based Risk Assessment for AI-enabled Biological Tools Centre for Long-Term Resilience · 2024 · Framework
- The Convergence of Artificial Intelligence and the Life Sciences Nuclear Threat Initiative · 2023 · Report
- Biosecurity Really: A Strategy for Victory Drew Endy et al. · Hoover Institution · 2025 · Strategy report
- SecureBio newsletters SecureBio · Newsletter collection
02
Biological capability & risk assessment
-
Global Risk Index for AI-enabled Biological Tools
Combines literature review and expert scoring of capability, accessibility and maturity to prioritise specialised tools for further assessment.
CLTR & RAND Europe · 2025 · Risk assessment -
Frontier AI Trends Report
Findings on biological knowledge, protocol generation and laboratory assistance, including comparisons with human baselines and evidence from user studies.
UK AI Security Institute · 2025 · Evaluation report -
ABC-Bench: An Agentic Bio-Capabilities Benchmark for Biosecurity
Tests tool-using biology agents on coding-heavy tasks. Accuracy excludes refusals; separate laboratory validation involved human assistance.
SecureBio · 2026 · Preprint
Further reading · 7
- Virology Capabilities Test (VCT): A Multimodal Virology Q&A Benchmark SecureBio-led research · 2025 · Preprint
- Why we recommend risk assessments over evaluations for AI-enabled biological tools (BTs) Centre for Long-Term Resilience · 2024 · Methods commentary
- Toward Comprehensive Benchmarking of the Biological Knowledge of Frontier Large Language Models RAND · 2025 · Research report
- Do the biorisk evaluations of AI labs actually measure the risk of developing bioweapons? Anson Ho & Arden Berg · Epoch AI · 2025 · Analysis
- Assessing the Impacts of Technology Maturity and Diffusion on Malicious Biological Agent Development Capabilities RAND · 2025 · Research report
- Evaluating Safety and Security in Biological Research RAND · 2025 · Perspective
- Preliminary Taxonomy of AI-Bio Safety Evaluations Frontier Model Forum · 2024 · Issue brief
- AI and Biotechnology Risks SecureBio · Research hub
03
Technical safeguards
Model safeguards, refusal behaviour and adversarial robustness.
-
BioTIER: A Refusal Benchmark for Targeted Biological Risk Mitigation
Measures refusal of hazardous biological requests and over-refusal of legitimate science using 542 expert-written prompts across 52 models.
SecureBio · 2026 · Preprint -
Constitutional Classifiers++: Efficient Production-Grade Defenses against Universal Jailbreaks
Assesses layered model safeguards through red-teaming and production-traffic evaluation, examining robustness, computational cost and false refusals.
Anthropic · 2026 · Preprint -
Boundary Point Jailbreaking of Black-Box LLMs
Adversarial testing of classifier-based safeguards, including biological-misuse evaluations, with vulnerabilities reported to developers for mitigation.
UK AI Security Institute · 2026 · Preprint
Further reading · 4
- Review of Anthropic’s Chemical and Biological Risk Report: Claude Opus 4.6 SecureBio · July 2026 · External assessment
- Principles for evaluating misuse safeguards of frontier AI systems UK AI Security Institute · 2025 · Evaluation framework
- Developing Guardrails for AI Biodesign Tools Nuclear Threat Initiative · 2024 · Paper
- Research Agenda for Safeguarding AI-Bio Capabilities Nuclear Threat Initiative · 2024 · Draft white paper
04
Synthesis screening
Sequence screening and customer verification at synthesis providers.
-
Beyond sequence similarity: toward function-based screening of nucleic acid synthesis
A framework for complementing sequence matching with prediction of biological function, addressing validation, false positives and the development of screening tools for AI-designed sequences.
Fourth Eon Bio · 2026 · Perspective -
Strengthening nucleic acid biosecurity screening against generative protein design tools
Tests screening against AI-designed sequences and documents coordinated software updates: an empirical example of safeguard assessment leading to deployed changes.
Wittmann et al. · Science · 2025 · Research paper -
A system capable of verifiably and privately screening global DNA synthesis
SecureDNA’s privacy-preserving screening architecture, with operational testing on commercial synthesis data and analysis of scalability and specificity.
Baum et al. · National Science Review · 2026 · Research paper
Further reading · 3
- Evaluating AI-assisted customer verification for synthetic nucleic acid screening Acelas et al. · Frontiers in Bioengineering and Biotechnology · 2026 · Research paper
- Why We’re Doubling Down on Synthesis Screening Sentinel Bio · 2026 · Prioritisation analysis
- UK screening guidance on synthetic nucleic acids for users and providers Department for Science, Innovation and Technology · 2024 · Guidance
05
Policy & governance
-
Defensive acceleration: the strategic pivot needed for UK biological resilience
Assesses 22 interventions and develops four programme proposals addressing surveillance, technology validation, automated screening and trusted access to frontier AI.
Centre for Long-Term Resilience · 2026 · Policy report -
Biological Tools and the EU AI Act
Applies general-purpose AI and systemic-risk definitions to 50 biological models, identifying ambiguities in the Act’s coverage.
Centre for Long-Term Resilience · 2025 · Policy analysis -
Biosecurity Governance Across Uncertain Artificial Intelligence Futures
Proceedings from a RAND–NTI workshop on risk thresholds, adaptive oversight and international coordination at the interface of AI and biosecurity.
RAND · 2025 · Conference proceedings
Further reading · 5
- Dual-use artificial intelligence and biology: upstream risk-benefit reviews Hanke et al. · Johns Hopkins Center for Health Security · Frontiers in Microbiology · 2026 · Hypothesis and Theory
- Risk-Based Categorization and Governance of Biological Data in AI Systems Allison Berke · Johns Hopkins Center for Health Security · 2026 · Report
- Priorities for the UK Government on frontier AI risk and Biosecurity Centre for Long-Term Resilience · 9 September 2026 · Policy brief
- How the UK Government should address the misuse risk from AI-enabled biological tools Centre for Long-Term Resilience · 2024 · Policy report
- America’s AI Action Plan The White House · 2025 · Government action plan
- Emerging Tech Policy Careers Horizon Institute for Public Service · Career guide
06
Physical biodefence
-
How I think about catastrophic biological risk — Part I
Examines respiratory protection and environmental biohardening against extreme biological threats, including circumstances where pathogen-specific countermeasures may arrive too late.
Andrew Snyder-Beattie · Defenses in Depth · 2026 · Strategic analysis -
Far-UVC (222 nm) efficiently inactivates an airborne pathogen in a room-sized chamber
Room-scale evidence for far-UVC air disinfection, measuring airborne microbial reduction rather than infection outcomes.
Eadie et al. · Scientific Reports · 2022 · Experimental study -
Innovate and Stockpile: Respiratory Protection for Essential Workers in a Catastrophic Pandemic
Examines respiratory protection requirements for catastrophic pandemics, with proposals for reusable equipment, product innovation and advance stockpiling.
Montazeri & Sandbrink · Health Security · 2023 · Commentary
07
Pathogen-agnostic surveillance
-
Inferring the sensitivity of wastewater metagenomic sequencing for early detection of viruses: a statistical modelling study
Combines empirical data from four studies to estimate detection sensitivity, linking sequence abundance with infection prevalence, incidence and sampling cost.
The Lancet Microbe · 2025 · Modelling study -
Metagenomic sequencing of composite airplane wastewater for surveillance of emerging viruses
Compares airplane and municipal wastewater at one airport and treatment plant, examining viral abundance and implications for surveillance sensitivity.
SecureBio-led research · 2026 · Preprint -
Building a three-day early-warning system for novel pathogens
A plan to shorten sample-to-result time from fourteen days to three through expanded sampling, sensitivity validation and operational changes.
SecureBio · 2026 · Programme update