Biosecurity Interventions (Overview)
Overview
The EA/x-risk community funds and operates a portfolio of biosecurity interventions designed to reduce catastrophic biological risks—both from natural pandemics and from deliberate misuse of advancing biotechnology and AI capabilities. Coefficient Giving alone has directed $230M+ toward biosecurity through 140+ grants.1
These interventions are organized around Kevin Esvelt's Delay, Detect, Defend framework, which structures biosecurity responses into three complementary layers: slowing dangerous capability proliferation, building early warning systems for novel threats, and developing resilience through countermeasures and physical defenses.
Delay / Detect / Defend Framework
Diagram (loading…)
flowchart TD
subgraph Delay["🛡️ Delay"]
DNS["DNA Synthesis Screening"]
AIR["AI Capability Restrictions"]
GOV["Biosecurity Governance"]
end
subgraph Detect["🔍 Detect"]
MET["Metagenomic Surveillance"]
BIO["AI Bio-Capability Evaluations"]
end
subgraph Defend["💊 Defend"]
MCM["Medical Countermeasures"]
UVC["Far-UVC & Physical Defenses"]
PPE["Next-Gen PPE"]
end
Delay --- Detect
Detect --- Defend- Delay interventions buy time by slowing the proliferation of dangerous capabilities—through DNA synthesis screening, AI output restrictions, and governance frameworks like the Biological Weapons Convention.
- Detect interventions build early warning systems. The most distinctive EA contribution here is untargeted metagenomic surveillance (sequencing all genetic material in wastewater samples), which can detect completely novel or engineered pathogens that would evade traditional PCR-based testing.
- Defend interventions build resilience so societies can survive even if delay and detection fail. This includes medical countermeasures (platform vaccines, broad-spectrum antivirals), far-UVC pathogen inactivation technology, and improved PPE.
The framework treats these layers as complementary: restrictions buy time while detection and defense infrastructure scales up.
Biodefense During Intelligence Explosion ("Crunch Time")
Ajeya Cotra has argued that biodefense should be among the top priorities for redirecting AI labor during the "crunch time" window --- the period between AI automating AI R&D and the arrival of vastly superhuman systems. During this window, AI capable of revolutionizing biodefense could also be exploited for offensive biological capability. Cotra's proposal is to use AI labor to build "much better infrastructure for monitoring, sequencing, detecting, and responding to biological threats" before superintelligent systems arrive, since bioweapons represent one of the most concrete catastrophic risks from AI capabilities even in scenarios where alignment is solved.
Intervention Categories
| Intervention | Key Actors | Description |
|---|---|---|
| DNA Synthesis Screening | SecureDNA, IBBIS/NTI, IGSC | Preventing orders of synthetic DNA that could reconstruct dangerous pathogens. SecureDNA provides free cryptographic screening to synthesis providers. |
| Metagenomic Surveillance | SecureBio/NAO, CDC | The Nucleic Acid Observatory sequences all genetic material in wastewater to detect novel threats. CDC's proposed Biothreat Radar ($52M) builds on this work.2 |
| AI Bio-Capability Evaluations | OpenAI | Testing whether AI systems can troubleshoot complex virology laboratory protocols, including the Virology Capabilities Test.3 |
| Medical Countermeasures | Red Queen Bio, BARDA, CEPI | "Defensive co-scaling"—pre-building countermeasures against AI-mapped biothreats. Platform vaccines and broad-spectrum antivirals. |
| Far-UVC & Physical Defenses | Blueprint Biosecurity, Columbia University | Far-UVC light (222nm) offers promise for continuous disinfection in occupied spaces and improved air cleaning. Blueprint runs the Airborne Infection Resilience (AIR) Program.4 |
| AI Capability Restrictions | Anthropic (ASL-3), OpenAI, Google DeepMind | Output filtering and deployment restrictions for frontier AI models based on CBRN capability evaluations. |
| Biosecurity Governance | NTI Bio, CSIS, Council on Strategic Risks | Strengthening the BWC, policy research on AI-enabled bioterrorism, dual-use research oversight.5 |
Key Organizations
| Organization | Focus | Key Funding |
|---|---|---|
| SecureBio | Delay/Detect/Defend framework; NAO; AI evals | $9.4M+ from Coefficient Giving |
| SecureDNA | DNA synthesis screening technology | Swiss foundation |
| IBBIS | International screening standards | Coefficient Giving via NTI |
| Blueprint Biosecurity | Far-UVC technology deployment | $900K from Coefficient Giving (2024) |
| Red Queen Bio | AI-driven medical countermeasures | $15M seed (OpenAI-led) |
| NTI Bio | Biosecurity governance, BWC | $7.8M from Coefficient Giving |
| Centre for Long-Term Resilience | UK biosecurity policy | EA-funded |
For a full list, see the Biosecurity Organizations section under Organizations in the sidebar.
Further Reading
- Is EA Biosecurity Work Limited to Restricting LLM Biological Use? — Detailed report analyzing the full intervention portfolio, funding flows, government adoption, and the restriction vs. resilience debate.
Sources
Footnotes
References
1The Virology Capabilities Test (VCT): A Benchmark for LLM Capabilities in Virology Laboratory ProtocolsarXiv·Jasper Götting et al.·2025·Paper▸
The Virology Capabilities Test (VCT) is a 322-question multimodal benchmark evaluating LLM capabilities in complex virology lab troubleshooting, created with PhD-level expert input. Remarkably, OpenAI's o3 model scores 43.8%, outperforming 94% of human expert virologists who average only 22.1%. The authors argue these dual-use capabilities warrant integration into biosecurity governance frameworks.