Anthropic says Claude was used to develop guided weapons in Yemen

Anthropic said a cell operating in northern Yemen used its Claude AI assistant to develop software for missile guidance and control, moving beyond theoretical work to a field launch test. The company said the activity, tracked as GTG-87001, involved coding, flight simulation, systems review and error analysis for guided weapons programs. Anthropic said the group ran several Claude instances at once and divided tasks between them like a small engineering team. The report described three weapons projects, including a guided missile with terminal guidance, a multi-stage ballistic missile project with a target range of more than 2,000 kilometers, and the R2000 missile designs, including a version with a hypersonic glide vehicle. Anthropic said many requests were blocked but not all, and that operators concealed the military purpose of some prompts across multiple sessions. The company said a field test of a guided missile inside Yemen likely failed, and that it has no evidence the cell brought an operational weapon into service. Anthropic said it banned the related accounts and shared information on the activity with public- and private-sector partners.