Report says Houthi-linked group tried using Claude for missile work

In short: Anthropic says it found and blocked a group in northern Yemen that tried to use its Claude chatbot to help with weapons development, but there is no evidence they built an operational missile.

What happened

Anthropic said it identified a cell in northern Yemen that was running three weapons development efforts. The company described them as a guided rocket program, a long range ballistic missile program, and a missile concept that included a hypersonic glide variant.

Anthropic said the users tried to use Claude to help write “guidance, navigation, and control” software. This is the code that helps a missile know where it is, where it is going, and how to adjust, like the combined role of a map app and autopilot for a weapon.

The company said the group carried out a test fire of a guided rocket that failed. Anthropic also said the group had already built an offline simulation toolkit before its accounts were blocked, meaning they had tools on their own computers to run practice tests in a virtual environment.

Separately, The Associated Press reported that the actors were in Houthi held areas of Yemen. It cited weapons researcher Trevor Ball, who said the Houthis may be trying to build more of their own capabilities so they rely less on shipments from Iran.

Why it matters

This reporting suggests some armed groups may try to use general purpose AI tools as one input in weapons work, rather than AI replacing outside support. It also shows the limits of what is known so far, including no evidence that the group fielded an operational missile or became independent of Iranian help.

Source: NYTimes