US artificial intelligence developer Anthropic has uncovered a troubling attempt by a threat actor cell based in northern Yemen to exploit its Claude AI platform to advance weapons development, including critical guidance, control, and navigation systems for ballistic missiles. The company detailed the incident in a Thursday report outlining actions it has taken to block unauthorized misuse of its AI technology across multiple countries.
Northern Yemen is the primary stronghold of the Houthi movement, which has controlled Yemen’s capital Sanaa since 2014 and just concluded a major military breakthrough this week. The group’s rapid offensive forced the retreat of forces aligned with Saudi Arabia and the internationally recognized Yemeni government headquartered in Aden, leaving the Houthis in control of large stretches of Yemen’s Red Sea coastline adjacent to the Arabian Gulf. This territorial gain includes key strategic islands and the Bab el-Mandeb Strait, a critical global shipping chokepoint that connects the Red Sea to the Gulf of Aden and borders three nations: Eritrea, Djibouti, and Yemen.
According to Anthropic’s report, the Yemeni cell was running three distinct weapons development initiatives when it turned to Claude for assistance. The programs included a guided rocket outfitted with an off-the-shelf phone-grade flight computer equipped with final-phase homing guidance technology, a multi-stage ballistic missile with an intended range exceeding 2,000 kilometers, and a flexible multi-variant missile platform that included a configuration for a hypersonic glide vehicle.
The threat actors leveraged Claude’s capabilities to develop critical software for these weapons projects, work that would normally require the expertise and labor of a full team of specialized engineers. Anthropic confirmed that while there is no conclusive evidence the group successfully deployed a fully operational weapon based on the AI-assisted work, the actors did carry out a test firing of the guided rocket. That field test ended in failure, and within hours of the unsuccessful attempt, the actors returned to the Claude platform to ask for help troubleshooting what went wrong.
As soon as the illicit activity was flagged and confirmed, Anthropic moved quickly to ban all accounts linked to the threat cell, cutting off their access to the AI system.
Public reporting has long documented that the Houthi movement receives arms supplies and military training from Iran, and a previous investigation by Middle East Eye also exposed additional military support from Russia. Despite this foreign backing, the Houthis have prioritized building up a domestic military industry to strengthen their independent defense and intelligence capabilities, a priority that has only grown amid years of ongoing conflict in Yemen.
Since Israel launched its full-scale invasion of Gaza in 2023, the Houthis have launched a sustained campaign of attacks on commercial shipping transiting the Red Sea, which the group says is an act of solidarity with Palestinians under siege in Gaza. A prior Middle East Eye report detailed how the group has used openly available commercial shipping data to target vessels linked to Israel and the United States for these attacks, marking a new pattern of asymmetric leveraging of accessible commercial and digital tools to advance military goals.