SAN FRANCISCO/SANAA: US artificial-intelligence company Anthropic says it disrupted a weapons-development operation based in northern Yemen that had been using its Claude AI system to assist with the design of guided rockets and missiles.
In a threat-intelligence report published on Thursday, Anthropic said the group was working on three separate weapons programmes: a guided rocket using a phone-class flight computer with terminal homing guidance, a multi-stage ballistic missile with a stated range exceeding 2,000 kilometres, and a missile family that included a hypersonic glide vehicle variant.
According to the company, the actors used Claude Code to develop guidance, navigation and control software, including flight-control firmware, position estimation, simulation and tuning. Anthropic said the group ran several Claude instances simultaneously, assigning them separate roles in a way that resembled a small engineering team.
Anthropic said it found no evidence that the group succeeded in fielding an operational weapon. It did, however, identify a live test of a guided rocket, which appears to have failed. Within hours of the test, the users reportedly returned to Claude to analyse what had gone wrong. The company subsequently banned the accounts linked to the operation and shared relevant information with public- and private-sector partners.
Anthropic did not identify the actors as Houthis, referring only to a cell based in northern Yemen. The disclosure comes as the Iran-backed Houthi movement expands its control along Yemen’s Red Sea coast and has seized the strategically important island of Mayun near the Bab el-Mandeb shipping route.
The Yemen case was one of several examples highlighted in Anthropic’s latest misuse report, which also described attempts by actors in Russia and China to use Claude for weapons-related software, cyber operations, intelligence gathering and other prohibited activities.