FT: Houthis attempted to develop ballistic missiles with the help of Anthropic's AI

Combatants in Yemen used the artificial intelligence model Claude—developed by the U.S. company Anthropic—in an attempt to create guidance, control, and navigation software for ballistic missiles, according to a threat report published by the company and cited by the Financial Times.

Anthropic identified a group linked to northern Yemen—a region largely controlled by the Houthis—that was developing three weapons programs. These included a multi-stage ballistic missile, a missile with multiple variants, and a guided rocket equipped with a flight computer based on a commercial smartphone.

The actors used Claude Code instead of human engineers to develop the programming designed to guide and stabilize the missiles during flight, according to Anthropic.

The company stated that its safety mechanisms blocked "many of their requests, but not all." The actors reportedly employed various tactics to evade restrictions, including concealing their objectives and the program's intended use. Furthermore, they broke the work down into multiple sessions to prevent a single interaction from fully revealing their intentions.

According to the report, the attempt to develop the missiles apparently failed, though it highlights the challenges of preventing advanced AI systems from being used for military purposes.