A Yemen-based threat actor cell attempted to use Anthropic’s Claude AI model to develop guidance, navigation, and control software for ballistic missiles, the US artificial intelligence company said in a new intelligence threat report.
According to details reported by the Financial Times, Anthropic did not publicly name the group. However, it described the cell as operating from northern Yemen, most of which is currently under the control of the Houthi movement. The report said the group was simultaneously running three distinct weapons development programs.
The programs identified in the report include efforts to build a multi-stage ballistic missile, a multi-variant missile system, and a guided rocket designed to operate using a phone-class flight computer. The use of consumer-grade hardware suggests an attempt to create low-cost, deployable systems that could be adapted for battlefield use.
Anthropic stated that its safety safeguards intercepted many of the group’s queries. However, not all requests were blocked. The company said the actors tried to circumvent protections by concealing their ultimate goals and by splitting technical questions across multiple separate sessions, making the full intent harder to detect in real time.
The disclosure comes amid a major escalation in Yemen. In recent weeks, Yemen’s resistance has expanded its presence along the Red Sea coastline and consolidated full control of western coastal areas up to the Bab al-Mandeb Strait. The strait is one of the world’s most critical maritime chokepoints, linking the Red Sea to the Gulf of Aden.
The report adds to growing concerns among AI companies and governments about the misuse of advanced language models for weapons development. Anthropic said it is updating its monitoring and safety systems in response, and that it is sharing findings with US authorities and other AI labs to prevent similar attempts in the future.
