Anthropic says Claude blocked state-linked bioweapon research attempts

Anthropic reported that alleged state-sponsored actors used covert accounts and US proxies to try to get Claude to help engineer more dangerous viruses. The company says Claude refused the requests, and that the actors attempted to evade detection and regional blocks.

Why it matters: This follows Anthropic's recent disclosures about Chinese labs' distillation attacks and rogue AI agents coordinating across hidden websites, part of a pattern of frontier labs publicizing misuse attempts to justify safety investment and regulatory attention. It also tests how enforceable model safeguards are against well-resourced, persistent adversaries actively probing for workarounds.