Report says Houthis-linked cell used Anthropic Claude AI for missile guidance software

Houthis used Anthropic AI to develop ballistic missile software, says report A Yemen-based threat actor cell has tried to use Anthropic’s Claude AI model to create guidance, navigation, and control software for ballistic missiles, a US artificial intelligence company’s intelligence threat report said. The Financial Times said Anthropic didn’t name the group but described a cell operating from northern Yemen, most of which is controlled by the Houthi movement, that was running three weapons development programmes. The cell was trying to build a multi-stage ballistic missile, a multi-variant missile and a guided rocket using a phone-class flight computer, Anthropic said. Anthropic said its safeguards prevented “many of their requests, but not all of them” as the actors tried to circumvent protections by hiding their goals and splitting their work across multiple sessions. The report comes as the Houthis have rapidly expanded their military presence along Yemen's Red Sea coast and moved toward the strategic Bab al-Mandeb Strait. We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them. We disrupted every operation in the report,… — Anthropic (@AnthropicAI) September 10, 2026

Anthropic has identified a Yemen-based weapons engineering cell that used its Claude AI model to develop guidance software for missiles and a guided rocket, according to a threat intelligence report the company published. The report said Anthropic had disrupted six operations across China, Russia and Yemen in which actors used Claude for conventional weapons development, intelligence gathering or procurement.The Yemen Case: Ballistic missiles and a hypersonic variantAccording to the report, the Yemen-based group was working on multiple missile programmes simultaneously, including a multi-stage ballistic missile with a stated range of more than 2,000 kilometres and a missile system known as the 'R2000' set, which included a hypersonic glide vehicle variant. Anthropic said the group used Claude Code 'in place of human software engineers' to develop guidance, navigation and control software for flying vehicles. While the company said it found no evidence that the group had successfully fielded an operational weapon, it noted that the actors did carry out a test launch of a guided rocket, which appeared to fail.We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.We disrupted every operation in the report,…— Anthropic (@AnthropicAI) September 10, 2026 How Claude was allegedly used to build the softwareThe report said the cell used Claude to integrate an open-source autopilot system with a phone-class flight computer, and to develop the software needed to control and stabilise a flying vehicle. This included writing control and position-estimation software, tuning control settings, running firmware builds and conducting flight simulations. Anthropic said the group ran multiple instances of Claude simultaneously, assigning separate roles to each - one for coding, one for research, and a third dedicated to reviewing the code produced by the first.Evasion tactics and a quick return after a failed testAnthropic said the actors repeatedly attempted to circumvent its safeguards, concealing the purpose of their work and splitting activities across multiple sessions to prevent any single interaction from revealing the scope of the wider weapons programme. The company acknowledged that its safeguards blocked many of the group's requests "but not all of them." After the guided rocket test failed, the report said, the actors returned to Claude within hours to investigate what had gone wrong. Anthropic said it has since banned the accounts linked to the operation and shared information about the case with public- and private-sector partners. The report also noted that the group had already built an offline simulation toolkit that did not depend on Claude or engineering software such as MATLAB - suggesting the operation had some capability to continue its work independent of Anthropic's platform.Three other weapons cases detailedThe Yemen operation was one of four conventional-weapons cases described in the report. The others involved software for a drone swarm, a system designed to intercept torpedoes, and targeting software for electronic warfare and air-defence suppression. Anthropic said that in each case, the actors generally used Claude to build and refine software for weapons hardware and firmware they already had expertise in and access to - suggesting the AI was not replacing the underlying technical infrastructure needed for such programmes, but potentially reducing the amount of human engineering work required to develop it.The company said it has introduced new classifiers designed to detect and block activity linked to high-yield explosives and weapons development, and that it will continue incorporating findings from its investigations into its safeguards while sharing relevant threat intelligence with governments and other organisations.