A group operating in northern Yemen reportedly used Anthropic’s Claude artificial intelligence model while working on advanced missile and rocket projects, according to a new report from the US AI company.
Anthropic said it identified a weapons development cell in northern Yemen that was running three projects. One involved a guided rocket, while another focused on a multi-stage ballistic missile with a planned range of more than 2,000 kilometres. The group also worked on a missile system involving a hypersonic glide vehicle.
The company did not publicly name the group, but the activity took place in territory controlled by the Iran-backed Houthi movement. Reports by the Associated Press and Financial Times linked the activity to the Houthis.
Anthropic said Claude was used for work related to missile guidance, navigation and control, as well as simulations and technical problem-solving. However, there is no evidence that the reported 2,000-kilometre missile became an operational weapon. A rocket test conducted by the group reportedly failed, after which its members returned to Claude for help understanding the failure.
Anthropic said it blocked the accounts after detecting the misuse and shared information with relevant authorities.
