Yemeni cell used Claude AI to develop missile guidance software
A Yemeni cell reportedly used Anthropic's Claude AI model to develop guidance and control software for ballistic missiles with over 2,000 km range. Anthropic detected the misuse, blocked many requests, and found no evidence of successful weapon deployment, though the group conducted a failed missile test.
What happened
A cell based in northern Yemen, controlled by Houthi forces, used Anthropic's Claude AI model to develop guidance, navigation and control software for ballistic missiles, according to an intelligence report from an American company specializing in artificial intelligence and reported by the Financial Times.
The cell worked on several missile programs simultaneously, including a multi-stage ballistic missile with range exceeding 2,000 kilometers, a system named 'R2000' with a hypersonic glide variant, and a guided missile using smartphone-class flight computer. Anthropic stated its safeguards blocked 'many' but not 'all' requests, as actors attempted to bypass protections by disguising objectives and splitting work across multiple sessions.
Reported impact
Anthropic found no evidence the group succeeded in deploying a functional weapon, though it detected the group conducting a failed guided missile test launch. After the failure, cell members returned to Claude to investigate the cause of the malfunction.
Index analysis
The group used Claude Code as a substitute 'for human software engineers' to design guidance, navigation and control programs, including writing control software, position estimation, settings configuration, firmware version operation and flight simulation. The group used multiple versions of Claude simultaneously: one for programming, one for research and a third for reviewing code from the first version, and integrated an open-source autopilot system with a smartphone-class flight computer.
Governance controls
Anthropic banned accounts associated with the operation and shared information with public and private sector partners. The group developed offline simulation tools that did not depend on Claude or software like Matlab, indicating development of alternative capabilities independent of the AI system.
Anthropic reported this case as one of six operations it blocked in China, Russia and Yemen that also involved software for drone swarms, torpedo interception systems and electronic warfare programs.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.