Yemen's Houthis reportedly tried to use Anthropic's AI to develop missiles
A threat report from AI company Anthropic says Yemen's Houthi movement attempted to use its Claude model to develop guidance systems for ballistic missiles. Many of the requests were blocked, but not all.
The name "Houthis" has resurfaced in international news in recent days. Officially known as "Ansar Allah" (Supporters of God), the group is a radical political-military movement and Shia Islamic religious movement based in Yemen. It controls a large part of Yemeni territory, is a party to the country's civil war, and has carried out attacks on ships in the Red Sea. The Houthis receive military and financial support from Iran and its allies, including the Lebanon-based Shia organization Hezbollah.
Attempts to exploit artificial intelligence
The threat posed by the Houthis has reportedly taken on a new dimension. It has emerged that representatives of the group attempted to use Claude, the AI model developed by Anthropic, to build guidance, control and navigation software for their ballistic missile programs. The Financial Times reported this, citing an Anthropic threat intelligence report.
According to the report, a threat actor cell was identified operating in northern Yemen, in territory controlled by the Houthis. The group is said to be working on three separate weapons programs: a guided missile using a mobile-phone-class flight computer with a terminal-phase target-seeking guidance system; a multi-stage ballistic missile with a declared range exceeding 2,000 kilometers; and a multi-variant missile kit, referred to as "R2000", which reportedly includes a hypersonic glide vehicle variant.
Not all requests were stopped
Anthropic stated that its safety systems blocked many of the requests submitted by Yemeni actors, but not all of them. The case underscores growing concerns that AI tools can be misused to advance military technology development, and that such misuse can only be partially prevented by developers' safeguards.
