Anthropic says accounts in northern Yemen tried to use its Claude AI system to help develop advanced weapons, including guided rocket and missile capabilities, before the company blocked them. The company says the users relied on Claude Code to build guidance, navigation, and control software, pursued multiple weapons programs including a hypersonic-style glide missile concept, and carried out a failed guided-rocket test before returning to Claude to troubleshoot it.
Why it matters: This is a real-world example of AI systems being used in attempted weapons development, not just general misuse or theory. It matters to AI providers, policymakers, and the public because it points to urgent pressure for monitoring, controls, and enforcement around dual-use AI access by sanctioned or militant actors.
Bruce Schneier
2026.09.14
97% relevant
This is the same underlying Anthropic disclosure and adds specific details quoted from the report: the actors were based in northern Yemen, ran three weapons programs including an R2000 set and a multi-stage ballistic missile, used multiple Claude Code instances in parallel for coding, research, and review, tested a guided rocket, and appeared to return to Claude after a failed field test.
Associated Press
2026.09.12
100% relevant
The article establishes a distinct event: Anthropic says it identified and blocked specific Claude accounts in Houthi-held Yemen after they used the service in an apparent attempt to build guided-weapons capabilities.
← Back to all stories