Anthropic Blocked Accounts After Attempt to Use Claude for Weapons Software
Anthropic describes a group from northern Yemen that reportedly used Claude Code to develop guidance, navigation and flight-control software. The company blocked the accounts but has no evidence that a functional weapon was deployed.
On September 10, Anthropic published a case involving misuse of its model in which a group from northern Yemen reportedly used Claude while working on three programs for weapons software. According to the company, the work involved developing guidance, navigation and flight-control systems. Anthropic blocked the accounts associated with the activity and said it was sharing information with public- and private-sector partners.
The company did not identify the actors as Houthis. It said only that they were a group from northern Yemen. The Associated Press confirmed that Anthropic had published the case and that the company localized the activity to a part of Yemen controlled by the Houthis. However, geographic localization alone is not public evidence of a direct link between the group and the movement.
Weapons Software Developed Using Claude
According to Anthropic’s report, the group used Claude Code to integrate an open-source autopilot, run simulations and analyze telemetry data. These activities involved software intended for guidance, navigation and flight control.
Anthropic also said the group launched a guided missile in a field test. According to the company’s assessment, the test was unsuccessful. The company also stated that it had no evidence of the deployment of a functional operational device.
The report therefore does not confirm the development or combat deployment of a functional weapon with Claude’s assistance. It documents a publicly reported attempt to use generative AI in work on software for conventional weapons; Anthropic blocked the relevant accounts in connection with the activity.
Unverified Attributions and Stated Goals
Some technical details, the actors’ identities and the extent of the hardware actually created are based mainly on Anthropic’s internal investigation. They have not been independently forensically verified in public. A direct connection between the group and the Houthis has likewise not been confirmed.
According to Anthropic, the group stated its goal was to develop a multistage ballistic missile with a range exceeding 2,000 kilometers. However, this is a stated intention, not a confirmed technical capability or evidence that such a system exists.
What to Watch
Further developments could lead to a more precise attribution of the activity to a specific actor if governments or security partners publish such information. It will also be important to see whether Anthropic provides additional technical or attribution-related evidence about the case.
The effectiveness of Anthropic’s new classifiers in blocking requests related to weapons development may also be monitored.
Sources
- Anthropic — Countering misuse of AI: September 2026 – The company’s primary claims about the activity of the group from northern Yemen, its use of Claude Code, the unsuccessful test, the absence of evidence of an operational device and the blocking of the accounts.
- Associated Press – Independently confirms the publication of Anthropic’s findings and reports that the activity involved users in an area of Yemen controlled by the Houthis.
Verified and updated: 09/12/2026 06:21
