Anthropic says Claude blocked missile and espionage plots, but breaches persist

Anthropic reports Claude helped thwart multiple malicious operations, from missile-guidance software in Yemen to state-backed cyber-espionage and covert influence campaigns. While internal safeguards blocked many requests, some slipped through, with operators using separate sessions to disguise goals; accounts were banned and threat intel shared with partners. Incidents are linked to Midnight Blizzard/APT29 (Russia), a China-affiliated student operation, and Iranian state-aligned campaigns, including Uyghur-target recruitment in Syria. Anthropic is investigating repeated breaches amid tense relations with the Pentagon (blacklisting) even as Claude is reportedly deployed in some military missions, highlighting ongoing debates over AI regulation and safety.