AI Misuse Alert: Anthropic Blocks Houthi-Linked Missile Software Development

Anthropic PBC disclosed that a “cell of threat actors” in northern Yemen used Claude AI models to write guidance, navigation and control (GNC) software for three weapons programs: a guided rocket with a phone-class flight computer, a multi-stage ballistic missile targeting >2,000 km, and a missile family that includes a hypersonic glide vehicle. The actors ran multiple Claude instances—coding, researching, reviewing—substituting a small engineering team. Anthropic’s systems blocked many requests, but the group succeeded in test-firing a guided rocket, which failed; they returned to Claude within hours to diagnose the failure. Anthropic banned the accounts and shared the intelligence with partners. The findings were released in a threat-intelligence report on September 10 and highlighted in a September 11 Reuters briefing.

Yemen missile cell used Claude to develop weapons software, Anthropic says TRT WorldClaude AI Misused in Alarming Houthi Missile Development Bid Houthis used Claude AI to help build missile guidance systems Anthropic says Yemen fighters used AI to build guided rockets and missiles New York PostTerrorists who launch attacks on US ships in Red Sea used Anthropic to try to build ballistic missiles Times of IsraelClaude, the weapon of war that Anthropic didn't want to create: How Iran, the Houthis, China and Russia tried to use it for military and espionage purposes

This is the third Anthropic report since March 2025 documenting illicit activity on its platform. Safeguards originally targeted novice bioweapon attempts, but newer models can assist with complex engineering, prompting stricter “dual-use” restrictions. The Yemen cell’s activity coincided with an intensified Houthi offensive that has seized Red Sea locations, including Bab el-Mandeb, raising global shipping concerns. While Anthropic did not name the Houthis, the cell’s location in Houthi-controlled northern Yemen links the actors to the Iran-backed movement.

Middle East Eye

• Guided Rocket – Commodity phone-class flight computer and final-phase homing guidance.

Yemen missile cell used Claude to develop weapons software, Anthropic says

• Multi-stage Ballistic Missile – Range goal >2,000 km (?1,250 mi).

The Independent

• R2000 / Multi-variant Missile – Includes a hypersonic glide vehicle variant.

Associated Press

Anthropic detected an offline simulation toolkit the group built, capable of running without Claude or commercial software such as MATLAB, indicating that blocking the AI alone would not erase the technical knowledge already acquired.

Claude AI Misused in Alarming Houthi Missile Development Bid

• Anthropic: The activity was “identified in its pilot stage” and “we do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket.”

Anthropic says Yemen fighters used AI to build guided rockets and missiles

Anthropic reports no evidence of a fielded weapon but notes the offline simulation toolkit could enable continued development independent of Claude. The newer Claude Fable 5 and Mythos-class models were not used in these cases, limiting insight into potential abuse of future, more powerful models. No independent verification of the failed rocket test exists beyond Anthropic’s internal logs.

TRT World

• “We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket,” — Anthropic researcher

Times of Israel

• “This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed.” — Anthropic researcher

Times of Israel

• “The cases we share here aren’t typical misuse, but rather examples of the most notable and novel threat activity we’ve identified to date,” — Anthropic researcher

Times of Israel