Caught Scientists Using Claude for Bioweapon Research

Anthropic has shut down the accounts of researchers using Claude for biological weapons development, including those conducting gain-of-function experiments on viruses. These cases were detected by internal safety classifiers. We don’t know who the violators are, but Anthropic says it’s for their own safety.

This is the first case of an AI lab publicly acknowledging that its cutting-edge model was used in biological research that could have been dangerous. It’s hard to distinguish these requests from normal scientific use.