As of mid-2025, LLMs may already be providing substantial uplift in several relevant areas of bioweapon development, perhaps doubling or tripling the likelihood of success.
Pessimistic
Source
Dario Amodei — The Adolescence of TechnologyDario Amodeidarioamodei.com
View all claims from this article Original Quote
As of mid-2025, our measurements show that LLMs may already be providing substantial uplift in several relevant areas, perhaps doubling or tripling the likelihood of success.
Outlook
PessimisticDoubling or tripling the likelihood of successful bioweapon development by unskilled actors would substantially increase the risk of catastrophic biological attacks.
Confidence
measured empirically, stated as current finding
Potential Data Sources for Validation
Anthropic's internal red-team evaluations and biosecurity benchmarks, potentially validated by third-party AI safety institutes, should show measurable uplift in bioweapon-relevant task completion rates attributable to LLM assistance.
About the Author
Dario Amodei
Dario Amodei is the co-founder and CEO of Anthropic, the AI safety company behind the Claude large language model series, and previously served as vice president of research at OpenAI.
Author Perspective
Across 22 tracked claims, Dario Amodei skews pessimistic.
36% optimistic64% pessimistic