Photo: Anastasija Vujic / iStock / Getty Images
A group of threat actors in northern Yemen, believed to be Iran-backed Houthi militants, attempted to use Anthropic’s Claude artificial intelligence models to develop ballistic missiles, according to the company’s latest report. Anthropic said it detected and blocked the group’s efforts to leverage Claude for three weapons development programs, including a guided rocket, a multi-stage ballistic missile with a range goal above 2,000 kilometers, and a hypersonic glide vehicle variant.
The cell reportedly used Claude to help create guidance, navigation, and control software for these airborne weapons, work that would typically require a team of trained engineers. According to Anthropic, while there is no evidence the actors succeeded in deploying a functional weapon, they did conduct a failed test-fire of a guided rocket and returned to Claude to troubleshoot the failure shortly afterward. Anthropic immediately banned the involved accounts and shared information about the attempted misuse with government authorities and industry partners.
The cell in Yemen employed multiple instances of Claude, assigning each to specific engineering tasks and obscuring their ultimate objective by splitting requests across sessions. While most prompts were blocked, some slipped through due to these tactics. Anthropic has since reinforced safeguards on its newer Claude models to prevent similar incidents and published the findings to encourage industry-wide vigilance and regulatory action.
Anthropic’s report also details broader misuse of its AI models in other regions, including Iranian state-aligned accounts that used Claude for propaganda, surveillance, and targeting recommendations against US naval forces, according to Iran International. The company has documented influence operations spanning Russia, Iran, Turkey, South Asia, Africa, and Europe, including the development of social media campaigns and cyber-espionage activities.
The company’s findings come amid ongoing Houthi offensives in Yemen, including the recent seizure of the Red Sea port city of Mocha, heightening concerns over the use of advanced technology by armed groups in active conflict zones. The report did not specifically name the Houthi movement, but the actors’ control of northern Yemen and their history of attacks on US naval forces strongly indicate their involvement.
The report’s publication follows the resignation of a senior Anthropic researcher who cited concerns over AI safety and the risk that powerful models could elude human control. The company stated, “We hope that the findings in this report will help other developers recognize similar patterns on their own platforms, give governments and civil society a clearer view of how emerging threats take shape, and strengthen collective defenses.”
As AI models grow more capable, experts and lawmakers are calling for stricter government regulation, warning that relying solely on companies to police their own technology may be insufficient to counter increasingly complex misuse.