Iran-backed Houthi rebels in Yemen tried to use Anthropic’s powerful Claude AI model to craft software for use in ballistic missiles, according to an alarming report released by the company.
Anthropic did not identify the Houthis by name – instead referencing “a cell of threat actors based in northern Yemen running three weapons development programs” using Claude. The reference almost certainly refers to the Houthis, who control the region.
They were trying to build a “guided rocket that used a commodity phone-class flight computer with final-phase homing guidance,” as well as a “multi-stage ballistic missile” and a “multi-variant missile … that included a hypersonic glide vehicle variant,” according to Anthropic’s report, which was published Thursday.
A Houthi fighter carries a mock missile as he marches during a campaign to mobilize more fighters amid an escalation of fighting on Thursday in Sana’a, Yemen. YAHYA ARHAB/EPA/Shutterstock
“The actors used Claude Code in place of human software engineers to develop the guidance, navigation, and control (GNC) software that steers and stabilizes a flying vehicle,” the company said.
To get around Anthropic’s built-in safeguards for the AI model, the Houthis tried “hiding their goals and the products the software was meant for,” according to the report. They also broke up their request across multiple sessions of use.
Anthropic said its safeguards “blocked many of their requests, but not all of them.”
“We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket,” the report said. “This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed.”
Anthropic’s report detailed several other shocking episodes in which bad actors attempted to misuse its models in ways that could’ve caused harm.
In one case, an Iran-linked group asked Anthropic’s model to compile information that could be used against US military targets, including records on American ship and aircraft movements as well as weaknesses in their communications systems.
Elsewhere, the company said it disrupted efforts by scientists to use its AI models for research that could have aided the construction of biological weapons.
