NEW DELHI: A weapons-development cell linked to the Iran-backed Shia Islamist terrorist group Houthis in northern Yemen used Anthropic’s Claude AI to develop guidance software for rockets and missiles, including a guided rocket that was test-fired, according to Anthropic’s latest threat intelligence report published on Thursday, September 10, 2026.
Anthropic said it identified a cell of threat actors based in northern Yemen that was running three weapons-development programmes. The company subsequently banned accounts associated with the cell.
The Houthis, an Iran-backed Shia Islamist terrorist group that controls large parts of northern Yemen, have developed and deployed various missiles and drones during the conflict in Yemen and their attacks in the Red Sea region.
Anthropic said the Houthi-linked cell used Claude Code to develop guidance, navigation and control software for airborne weapons. The use of Claude allowed the actors to carry out technical work that would otherwise have required a team of software engineers. The case illustrates how artificial intelligence could potentially reduce the technical resources required by terrorist organisations and other non-state actors seeking to develop sophisticated missiles and guided weapons.
Houthi-linked cell worked on three missile programmes
According to Anthropic, the Houthi-linked weapons cell was simultaneously involved in three separate weapons-development programmes. The first involved a guided rocket equipped with a commercial, phone-class flight computer and final-phase homing guidance. The second involved a multi-stage ballistic missile with a stated range of more than 2,000 kilometres. The third was a multi-variant missile programme that included a hypersonic glide vehicle.
The Houthi-linked actors used several Claude instances at the same time, Anthropic said. Different Claude instances were assigned different functions, including coding, research and reviewing software generated by other instances. The AI was used to integrate an open-source autopilot with a phone-class flight computer, develop control and position-estimation software, tune flight controls, run firmware builds and conduct flight simulations.
The weapons cell also used Claude during the testing and troubleshooting process. Anthropic said the actors conducted a test-fire of the guided rocket. The test appeared to fail, and within hours the Houthi-linked actors returned to Claude to investigate the cause of the failure. Anthropic said it found no evidence that the group had successfully fielded an operational weapon, despite the guided-rocket test.
The incident demonstrates the potential role of AI in shortening parts of the weapons-development cycle. Instead of relying entirely on conventional engineering teams, the Houthi-linked cell used Claude for multiple technical functions, from software development and simulation to analysing problems following a rocket test.
Anthropic safeguards blocked some weapons requests
Anthropic said its safeguards blocked many requests from the Houthi-linked weapons cell, but they did not prevent all of the activity. The actors attempted to conceal the nature of their work by disguising their objectives and dividing activities across multiple Claude sessions. By separating individual tasks, the overall weapons programme was less apparent from any single interaction with the AI system.
The company said it subsequently banned accounts associated with the Houthi-linked cell. Anthropic also said it had no evidence that the group successfully fielded an operational weapon. Nevertheless, the test-fired guided rocket showed that the use of Claude had extended beyond theoretical research into practical stages of weapons development.
Anthropic’s disclosure is significant because the Houthis have previously demonstrated capabilities involving ballistic missiles, cruise missiles and drones. The report adds a new dimension to concerns surrounding the terrorist group’s technological capabilities by showing how commercially available AI systems could be incorporated into weapons-related software development.
