Anthropic Disrupts Yemen-Based Group Using Claude AI for Missiles

AI Company Uncovers Weapons Development in Yemen

Anthropic has disclosed that it detected and disrupted a group operating in northern Yemen that was using its Claude AI model to develop guidance systems for missiles and rockets. The company detailed the incident in its latest report on AI misuse detection, published Thursday.

The unidentified actors were working on three separate weapons projects: a guided rocket, a ballistic missile with a stated range exceeding 2,000 kilometers, and a missile system designated as the R2000. They used Claude to assist with creating guidance, navigation, and control software—tasks that would typically require specialized software engineering expertise.

According to Anthropic, the group attempted to evade detection by distributing their work across multiple AI sessions and concealing the military purpose of their projects. The actors conducted at least one live rocket test and subsequently consulted Claude to troubleshoot what appeared to be a launch failure.

Technical Sophistication and Offline Capabilities

The Yemen-based group demonstrated notable technical planning by developing offline simulation toolkits. These tools would allow weapons development work to continue even without continuous access to Claude or standard engineering software environments, suggesting an awareness of potential access disruptions.

Anthropic emphasized that its investigation found no evidence the effort produced an operational weapon system.

Why It Matters

This incident represents one of the first documented cases of non-state actors attempting to use commercial AI systems for weapons development. It underscores a critical challenge facing AI companies: preventing misuse while maintaining useful capabilities for legitimate engineering and research. The disclosure also arrives amid heightened tensions in the Red Sea region, where Iran-backed Houthi forces have been conducting missile and drone attacks against commercial shipping along a vital global trade route. Recent fighting between Houthi militants and Saudi-backed Yemeni forces near the Bab el-Mandeb Strait has intensified, contributing to volatility in energy markets.

Company Response and Safeguards

Following the investigation, Anthropic took several actions: it banned all accounts connected to the activity, shared intelligence with partner organizations, and implemented enhanced safeguards to prevent similar misuse. The company has not disclosed specific details about the strengthened detection mechanisms.

The case highlights the growing need for AI companies to develop robust monitoring systems capable of identifying dual-use applications of their technology, particularly when users deliberately obscure their intentions.

This report was first published by Bloomberg, with reporting by Prinsloo and El Chmouri.