Threat actors used Claude Code to develop guided weapons software, Anthropic reports

Anthropic said a cell of threat actors based in northern Yemen used its Claude Code models to develop guidance, navigation, and control (GNC) software for three weapons development programs. The programs included a guided rocket with a commodity phone‑class flight computer and final‑phase homing guidance, a multi‑stage ballistic missile with a stated range goal above 2,000 km, and a multi‑variant missile set (referred to as the “R2000” set) that included a hypersonic glide vehicle variant.

The actors used multiple Claude instances at once, assigning each a role similar to a small engineering team: one instance wrote code, another performed research, and a third reviewed the code produced by the first. Anthropic said its safeguards blocked many of the actors’ requests but not all, and that the actors employed tactics to evade detection, including hiding their goals and splitting their work across multiple sessions so no single session revealed their full intent.

Anthropic reported that the actors carried out a sustained effort to develop guided weapons, including using Claude to design guidance software, and that the actors test‑fired a guided rocket. The company said the test appears to have failed and that within hours the actors returned to Claude to troubleshoot the failure.

Anthropic stated it does not have evidence the actors succeeded in fielding an operational device. The company emphasized that the actors’ approach demonstrates how AI systems can democratize expertise and capability, which can be beneficial but also poses risks when misused for harmful ends.