English

SecurityAnthropicClaude Code

Group Likely Linked to Houthis Used Claude Code for Missile Guidance Software Development

A cell in northern Yemen, highly likely to be linked to the Houthis, used Anthropic’s Claude Code to perform engineering tasks typically requiring a specialized team. According to Anthropic's September threat report, the operators used the model to develop software for guided rockets, a long-range ballistic missile, and a hypersonic glide vehicle concept known as “R2000.”

The group operated multiple Claude sessions in parallel to divide tasks between coding, research, and technical review. Their workflow included writing navigation and control code, performing reinforcement learning to tune flight-control algorithms, and conducting six-degree-of-freedom trajectory simulations to model movement and rotation in space.

The use of AI extended into a complete engineering cycle. After a test-fire of a guided rocket in Yemen failed, the operators reportedly used Claude to analyze launch telemetry to investigate the failure. While Anthropic found no evidence that an operational weapon was successfully fielded, the group had already compiled their work into a standalone executable that could run offline, reducing their dependence on direct access to the model.

To bypass safety safeguards, the operators reportedly obscured the intended end use of individual tasks and fragmented their work across separate conversations. Anthropic has since banned the accounts linked to this activity.

Sources

  1. Houthis Used Claude Code to Develop Missile Guidance Software: Anthropic (Hacker News Frontpage, 2026-09-13)