Noah Smith
· Noahpinion
· 28 августа 2026, 09:48
· ⏱ 7 мин чтения
| Читать в Substack ↗
Резюме
Noah Smith argues that AI-enabled bioterror — specifically a jailbroken LLM designing and procuring engineered supervirus strains — is a near-term, civilization-level risk that deserves far more attention than climate change or other AI fears. For markets, this is a macro tail-risk argument with no directly tradeable instrument in the article; it implies potential future policy, biosecurity, and defense spending rather than any current position.
•Smith assigns AI-enabled bioterror roughly a 10% chance of ending civilization and a 30% chance of causing truly world-changing destruction.
•His central scenario: by 2029, a disaffected teenager uses a jailbroken LLM to design 100 contagious viruses with long asymptomatic periods and high mortality, then coordinates an unlicensed biolab to ship samples.
•He argues vaccines cannot save humanity because asymptomatic spread gives the virus first-mover advantage; even fast AI-designed vaccines would face a slow physical rollout.
•He claims antivirals might save only about 5% of people, plus another 5% who are naturally resistant — not enough to sustain civilization.
•He believes AI bio risk is bigger than climate change: more catastrophic, nearer-term, and with less known about how to counter it.
•He criticizes the AI safety community for focusing on preventing superintelligence rather than on concrete bioterror pathways.