Anthropic Exposes AI Misuse for Bioweapons and Missile Systems

AI developer Anthropic PBC published a 154-page threat intelligence report on September 10, 2026, detailing how state-aligned actors and cybercriminals misused its Claude AI models between December 2025 and August 2026. The company identified attempts to develop biological weapons, including research to make the chikungunya virus more contagious and evade human immunity. Anthropic noted that while it disrupted these operations, it can no longer guarantee that its most capable models will not assist skilled actors in dangerous biological work.

In northern Yemen, a cell likely linked to the Houthi rebels used Claude Code to develop guidance and control software for a guided rocket, a hypersonic glide vehicle, and a multi-stage ballistic missile with a range exceeding 2,000 kilometers. While the group test-fired a guided rocket, Anthropic reported the test failed. Simultaneously, actors linked to the Government of Iran used the AI to create targeting handbooks against United States Navy assets by synthesizing satellite imagery and transponder data. Other military misuses included Russian freelance developers programming autonomous kamikaze drones and Chinese actors drafting proposals for anti-torpedo fire control systems.

Beyond weaponry, the report detailed mass surveillance and espionage. A consultant for the State Intelligence Service of Mali deployed a system to monitor 25 million mobile phones, while Chinese operations targeted Uyghurs and dissidents in Hong Kong. Anthropic also accused Chinese firms Moonshot and DeepSeek of distilling its technology via fraudulent accounts. These disclosures follow the resignation of former researcher Jacob Coxon, who accused AI companies of gambling with human lives in a race toward superintelligence.