Biological Threat Exposure

Biological Threat Exposure measures how AI systems increase the risk of catastrophic biological attacks by democratizing dangerous knowledge. The core concern is that AI makes capabilities previously requiring rare expertise accessible to a broader range of actors—the AI Uplift Assessment Model estimates current AI provides 1.3-2.5x uplift for novice actors.

Empirical evidence remains contested. RAND’s 2024 study found no significant AI uplift, while Microsoft found AI-designed toxins evaded 75%+ of DNA synthesis screening. In 2025, both OpenAI and Anthropic flagged biological capabilities as reaching concerning thresholds in their frontier models.

Metric	Score	Notes
Changeability	45	Moderately changeable through biosecurity measures and model restrictions
X-risk Impact	80	High—one of the most direct pathways to catastrophic harm
Trajectory Impact	40	Primarily affects near-term risk levels
Uncertainty	60	Substantial uncertainty around actual uplift magnitude

Related Content

Risks:

• AI-Enabled Bioweapons

Models:

• Bioweapons AI Uplift
• Bioweapons Attack Chain
• Bioweapons Timeline

Key Debates:

• How much does AI actually help with bioweapons—marginal or transformative?
• Does AI help biodefense more than bioattack, or vice versa?
• Are wet lab skills the real bottleneck, making AI uplift less relevant?

What links here

• Human-Caused Catastrophescenariokey-factor
• Misuse Potentialrisk-factorcomposed-of
• Bioweapons Attack Chain Modelmodelmodels
• AI Uplift Assessment Modelmodelaffects

---

Ratings

Metric	Score	Interpretation
Changeability	45/100	Somewhat influenceable
X-risk Impact	80/100	Substantial extinction risk
Trajectory Impact	40/100	Significant effect on long-term welfare
Uncertainty	60/100	Moderate uncertainty in estimates