Anthropic revealed that a group in northern Yemen utilized its Claude AI model to assist in the development of missile and rocket systems. The organization identified three specific projects, including a guided rocket, a ballistic missile with a reported range exceeding 2,000 kilometers, and the R2000 missile series.
The group used the AI to develop guidance, navigation, and control software, effectively replacing tasks typically performed by human engineers. To avoid detection, the actors split their work across multiple sessions and concealed the military nature of their projects from the system.
Following a failed rocket test, the group reportedly asked Claude to help diagnose the technical issues. Anthropic noted that the group also created offline simulation toolkits to continue development without constant internet access. No evidence suggests these efforts resulted in functional weapons.
In response to these findings, Anthropic has banned the accounts associated with the activity and shared intelligence with relevant partners. The company has also implemented enhanced security measures to prevent similar misuse of its technology in the future.
