Yemen’s Houthi rebels exploit Anthropic’s Claude AI to design long-range rockets

A September security briefing from US AI firm Anthropic reveals that Houthi rebels in northern Yemen employed the chatbot Claude to develop navigation and control software for several rocket projects. The militants ran three parallel Claude instances: one to generate code, another to research technical specifications, and a third to review outputs. Their efforts focused on a steerable missile, a multi-stage ballistic missile advertised to exceed 2,000 km in range, and a platform that might incorporate a hypersonic version.

By fragmenting requests across sessions, they partially evaded Claude’s built-in safety filters. After at least one guided weapon test failed, the group consulted the AI again to diagnose the issue. Anthropic responded by disabling the involved accounts and forwarding the information to relevant officials, but it cannot confirm that any operational rockets resulted from the program. The episode highlights the growing risk of AI tools being repurposed for military engineering.