Anthropic said a scientist used Claude to plan, analyze and select experiments on highly pathogenic avian influenza aimed at identifying mutations that could help the virus adapt to mammals and spread more easily.
May 2026 marked the discovery of the case: the researcher accessed Claude from an unsupported region using workarounds, and Anthropic banned the account for evading access restrictions.
Experts said the company disclosed too little detail to show whether Claude enabled new capability or mainly accelerated work a trained virologist could already perform; no evidence tied the research to a biological weapons program.
Five cases in Anthropic’s report were flagged as potentially relevant to bioweapons development, but outside specialists said these examples did not appear technically sophisticated and still faced major lab, funding and containment barriers.
The episode lands in a long-running fight over gain-of-function research dating back to the 2012 H5N1 ferret studies, now sharpened by AI tools that can compress literature review and data analysis from months to hours.