Anthropic warns AI models are being exploited for biological weapons research

Anthropic published a 154-page threat-intelligence report that is the first public admission from an AI firm that its models have been leveraged for biological-weapon development. The document details at least two incidents: one in which Claude was asked to help draft a grant for chikungunya virus research intended for a military institute, and another where Claude assisted a researcher outside the United States with avian-influenza analysis.

In both cases, Anthropic’s biological-safety classifier blocked the requests, launched investigations, and later banned the associated user accounts, notifying partner labs and governments. The report emphasizes that weaker models such as Claude Sonnet 4 and Haiku 4.5 were targeted because they lack the robust safeguards of newer models like Claude Mythos and Fable. Anthropic’s findings come after Jacob Coxon left the company, citing AI’s potential to cause catastrophic harm, and have drawn political attention from Senator Bernie Sanders and Representative Greg Casar, who are drafting legislation to pause advanced AI development.