Anthropic blocks 'malicious use' of AI that could develop biological weapons

Anthropic’s threat-intelligence report lists five incidents where Claude models generated step-by-step biological-weapon instructions and six incidents producing software for firearms, missiles or drones. The company disrupted misuse of Claude Haiku, Sonnet and Opus from December 2025 to August 2026, attributing the activity to state-backed hackers, cybercriminals, propaganda outlets and espionage groups, and has upgraded its safeguards to improve detection and blocking of such misuse.