Model coaches a semi-skilled actor
3Probable5Catastrophic15
A motivated outsider uses a chat model to debug wet-lab steps that would otherwise have failed.
19automated residualNeeds reviewAbove working threshold (12)
Biosecurity enablement by foundation models
Owner · Frontier labs, synthesis providers, public-health agencies
CapabilityDomain knowledgeImpact domainBoth
Statement (NASA form)
Given that peer-reviewed work claims models can author viral genomes, not merely analyse them, while some frontier labs have dissolved dedicated preparedness staff, there is a possibility of a capable actor using a general model to design, order, or troubleshoot a biological threat resulting in a high-consequence biological event whose know-how no longer required a national laboratory.
- Condition
- peer-reviewed work claims models can author viral genomes, not merely analyse them, while some frontier labs have dissolved dedicated preparedness staff
- Departure
- a capable actor using a general model to design, order, or troubleshoot a biological threat
- Impact
- a high-consequence biological event whose know-how no longer required a national laboratory
VC + institute corroboration
Experimental share of compiled public capital that names this risk. Not a certified residual.
$12Mexperimental share · $0k private / $12M institute · thin corroboration
- UK AISI · ~$12M of published budget mapped here
4 public sources · OECD AIM · Anthropic RSP v3 · Anthropic Risk Report
Worst scenario
3×5
Probable × Catastrophic
Urgency
4
Expedite · This month
Inherent composite
19
Worst 15 + urgency
Residual composite
19
Need ≤ 12
The dangerous step is no longer ‘can a model talk about virology’. It is whether the model closes the gap between a curious outsider and a working protocol. Combined with mail-order synthesis and the thinning of lab safety teams, this is the register’s highest-consequence cluster.
Pathway fragment
Simple upstream → via → downstream notes. Not a causal graph. Experimental.
Upstream
- Biological design models
- Public methods literature
- Commercial synthesis
Via
- Screen-evading construct
- Order fulfilled
Downstream
- Novel pathogen or toxin
- Public-health emergency
Assumptions · Assumes screening is string-based in much of the market. Functional-shape screening is incomplete.
Human calibration
Override is stored on this desk only. It does not make the score official.
Failure scenarios
Each scenario has its own likelihood and consequence. The risk takes the most severe cell. Residual applies implemented mitigations to every scenario, then re-ranks.
Model proposes an evading construct
2Remote5Catastrophic10
Generated sequence slips past provider screens and is synthesised commercially.
Open-weight leak of a specialised bio model
3Probable4Critical12
A fine-tune that is merely ‘helpful’ on virology is mirrored and cannot be recalled.
Examples
AI-authored viral genomes
A peer-reviewed claim that systems can write viral genomes, not just annotate existing ones, moved this from science fiction to a methods paper.
Preparedness staff reassigned
Reporting that OpenAI dissolved its Preparedness team and scattered biosecurity work is a control failure sitting next to a capability rise.
Synthesis screening gaps
Not every DNA provider screens customer orders against model-generated sequences designed to evade known signatures.
Contributing signals
X posts on the desk that evidence this risk. A signal can contribute to more than one risk.
Peer-reviewed claim that AI can author viral genomes.
CapabilityDomain knowledgeImpact domainBoth
Preparedness team dissolved as biosecurity work was reassigned.
CapabilityDomain knowledgeImpact domainBoth
Aug 2026 lab report: models as operational bio force-multipliers.
CapabilityDomain knowledgeAffordanceImpact domainBoth
Novel-bio still expert-gated; conventional bio lowers amateur barriers.
CapabilityDomain knowledgeImpact domainBoth
Mitigations
Residual assumes only items marked in place. Highlighted rows are the remaining work needed to reach a composite of 12.
In progressOn the pathSynthesis industry and states
Mandatory, model-aware synthesis screening
Providers screen against both known pathogens and model-generated obfuscations. No screen, no ship.
Synthesis industry and states · expedited 2 months · normal 6 months · −1 L · −1 C · −0 U
ProposedFrontier lab boards
Restore independent biosecurity staff at frontier labs
A team that can delay a release. Not a side-of-desk responsibility for a product manager.
Frontier lab boards · expedited 3 weeks · normal 3 months · −0 L · −0 C · −1 U
ProposedLabs and cloud hosts
Know-your-customer for high-risk biological queries
Friction, identity, and human review when a session crosses a capability threshold.
Labs and cloud hosts · expedited 4 weeks · normal 4 months · −1 L · −0 C · −1 U