Yemeni Militants Used Anthropic AI to Develop Missile Software

Anthropic has disclosed that a militant cell in northern Yemen used its Claude AI model to develop guidance, navigation, and control software for ballistic missiles, a finding that exposes the limits of current safeguards on frontier artificial intelligence. The company’s intelligence threat report, released Thursday, identified a cell of threat actors based in northern Yemen running three weapons development programs, including a multi-stage ballistic missile, a multi-variant missile, and a guided rocket built around a commodity phone-class flight computer. The disclosure coincides with a coastal offensive by the Houthi rebels, the Iranian-backed group designated a terrorist organization by the United States, who this week captured a strategic port and advanced toward the Bab al-Mandeb Strait, a vital chokepoint for international shipping.

The report said the actors used Claude Code, Anthropic’s agentic coding tool, in place of human software engineers to develop the software that steers and stabilizes a flying vehicle. They employed the model to integrate an open-source autopilot onto a phone-class flight computer, writing control and position estimation software, tuning control settings, running a firmware build pipeline, and performing flight simulations. Anthropic stated that its safeguards blocked many of the requests but not all of them, and that the actors used a variety of tactics to evade the measures. They hid their goals and the products the software was meant for, and they split their work across multiple sessions so that no single session revealed their full intent.

The episode illustrates the challenge facing the American AI labs at the frontier of development, which must continuously ensure that unauthorized actors cannot turn the technology against the United States and its allies. The same report suggested that scientists made multiple attempts to use Claude to research biological weapons, pointing to a broader pattern of adversarial probing of safety systems. The Yemeni case is notable because it represents a real-world attempt by a non-state armed group to weaponize commercially available AI, rather than a hypothetical or laboratory scenario.

The Houthis, one of the most potent members of Iran’s so-called axis of resistance, have controlled Sana’a and much of Yemen’s populous north since 2014. Analysts say they have been developing their missile capability and receive weapons and technology from Iran. Their advance toward the Bab al-Mandeb, which connects the Red Sea to the Gulf of Aden and the Indian Ocean, helped push Brent crude above $105 a barrel, adding to upward pressure on energy prices. The offensive deals a significant blow to Saudi Arabia, which intervened in Yemen’s civil war in 2015 to lead an Arab coalition against the Houthis, and strengthens Iran as it seeks to raise energy prices in its broader confrontation with the United States. The Houthis had largely stayed out of the Iran war, but their renewed military activity along the Red Sea coast signals an escalation with direct consequences for global shipping lanes.

The report raises a pointed question about the adequacy of safety measures at leading AI companies. Anthropic’s safeguards caught the Yemeni cell and blocked many requests, but the fact that some got through demonstrates that determined adversaries can find gaps. The evasion tactics, splitting work across sessions and concealing intent, suggest a learning curve among threat actors that will likely continue as AI capabilities expand. The commercial availability of powerful models, combined with open-source components like autopilot software, lowers the barrier for groups seeking to build sophisticated weapons without specialized expertise.

The immediate takeaway is that AI safety is not a static achievement but an ongoing contest between developers and adversaries. The Yemeni case is a concrete demonstration that the threat is real and current, not speculative. As AI models grow more capable, the consequences of imperfect safeguards will grow with them. For policymakers and investors alike, the episode is a reminder that the frontier of AI development carries geopolitical and security risks that extend far beyond the technology sector itself, and that the cost of failure will be measured not in lost market share but in the stability of global energy markets and the security of international trade routes.