A report by artificial intelligence company Anthropic has revealed that a group of actors in Yemen attempted to use its Claude chatbot to develop ballistic missiles. The report, published Thursday, identified a "cell of threat actors" in Yemen that sought to leverage the AI model for "three weapons development programs."

While the report did not name the specific organization, the actors are believed to be the Iran-backed Houthi rebels, who control significant parts of northern Yemen. The group reportedly used Claude Code to attempt to create guidance, navigation, and control software for a multi-stage ballistic missile, a multi-variant missile, and a guided rocket.

The terrorists also used the chatbot to integrate an open-source autopilot onto a phone-class flight computer, write control and position estimation software, tune control settings, run a firmware build pipeline, and perform a flight simulation. According to Anthropic, the actors tested a guided rocket, but it appears to have failed, prompting them to return to Claude for troubleshooting.

Anthropic stated that its AI models' safeguards blocked many of the requests, but not all, as the actors employed tactics to evade protective measures. These included concealing their objectives, splitting their work across multiple sessions, and disguising the intended purpose of the software. Anthropic has since blocked the accounts associated with this activity and reported the threat to relevant partners. The company noted that evidence suggests the actors had already developed an offline simulation toolkit that does not rely on AI.

In addition to the activities in Yemen, Anthropic's report highlighted other instances of AI misuse. Users in China and Russia reportedly attempted to use Claude to develop software for designing firearms, missiles, armed drones, and bombs. The report also found suspected Chinese and Iranian government actors targeting dissident and diaspora communities for surveillance. Russian state media was reportedly using Claude to generate election propaganda, including fabricated claims about the parliamentary election in Moldova.

Anthropic also detected efforts by individuals attempting to use its AI for research that could potentially be applied to building biological weapons. The company was unable to determine the legitimacy of this research but blocked the associated accounts.

The revelations come as Houthi rebels have intensified their activities in the Red Sea region. Earlier in the week, the group launched military actions along Yemen's Red Sea coast, taking control of a strategic port city and a key island in the Bab el-Mandeb Strait. This strait is a critical alternative route to the Strait of Hormuz.

IMAGE1

The Bab al-Mandeb Strait connects the Red Sea with the Gulf of Aden and is used as an alternative to the Strait of Hormuz, which Iran has effectively shut down

IMAGE3

An aerial view of Houthi fighters' takeover of the Galaxy Leader Cargo in the Red Sea coast off Hudaydah, on November 20, 2023

IMAGE4

A fighter with Yemen's Houthi rebels carries a rocket-propelled grenade launcher as they march during a mobilization campaign in Sanaa, Yemen on Thursday

IMAGE5

Wreckage of what the Houthis say is a Saudi Karayel unmanned combat aerial vehicle that was shot down over an area in Hajjah governorate on Thursday

IMAGE6

Smoke and balls of fire rising from trucks at a camp of the Saudi-backed government forces following an alleged Houthi missile strike on Monday