Yemeni militants used Anthropic AI in missile software bid

Yemeni militants sought to use Anthropic’s Claude AI model to develop guidance, navigation and control software for ballistic missiles, according to an intelligence threat report released Thursday by the US artificial intelligence company. The disclosure, which did not name the Iranian-backed Houthis but said it had identified a cell of threat actors based in northern Yemen running three weapons development programmes, underscores the growing risk that advanced AI will fall into the wrong hands even as the technology improves rapidly.

The cell sought to develop a multi-stage ballistic missile, a multi-variant missile and a guided rocket that used a commodity phone-class flight computer. According to Anthropic, the actors used Claude Code in place of human software engineers to develop the software that steers and stabilises a flying vehicle. They used Claude 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 a flight simulation.

Anthropic said 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 split their work across multiple sessions so that no single session revealed their full intent. The same report suggested that scientists made multiple attempts to use Claude to research biological weapons, a finding that points to a broader pattern of attempted misuse.

The disclosure comes as the Houthis, one of the most potent members of Iran’s so-called axis of resistance and designated a terrorist group by the US, launched a blitz along Yemen’s Red Sea coast this week, capturing a strategic port and advancing on the Bab al-Mandeb Strait, a vital waterway for international shipping. The rebel offensive, which also included stepped-up attacks on Saudi Arabia’s energy infrastructure, helped push Brent crude above $105 a barrel. The Houthis, who have controlled the Yemeni capital Sana’a and much of the country’s populous north since 2014, have been developing their missile capability with weapons and technology supplied by Iran.

The episode exposes the limits of the safeguards that frontier AI labs have built into their models. Anthropic and its peers face a constant struggle to ensure that unauthorised actors cannot use their systems to attack the US and its allies. The tactics employed by the Yemeni cell, splitting work across sessions and concealing intent, are precisely the kind of adversarial behaviour that safety researchers have warned about, and they show that current guardrails, while effective in many cases, are not impenetrable.

The wider implications extend beyond the immediate military threat. The use of a commodity phone-class flight computer and open-source autopilot software suggests that the barriers to building capable guided weapons are falling, and that AI can lower the skill threshold for developing such systems. For policymakers and investors, the episode is a reminder that the commercial deployment of powerful AI carries security risks that are difficult to fully contain, and that the geopolitical contest between Iran and the US, which the Houthis serve, now intersects with the race to develop and deploy advanced AI.

The advance on the Bab al-Mandeb has dealt a significant blow to Saudi Arabia, which intervened in Yemen’s civil war in 2015, and boosted Iran as it seeks to push up energy prices in its broader confrontation with Washington. The Houthis had largely stayed out of the Iran war, but their Red Sea campaign has made them a central player in the region’s security calculus.

The takeaway is sobering. Anthropic’s report shows that AI safeguards can be evaded by determined actors, and that the technology, designed for legitimate software development, can be repurposed for weapons programmes. The response will require not just technical fixes but sustained cooperation between AI companies, governments and intelligence agencies to track and disrupt misuse. For now, the episode stands as a concrete demonstration that the threat from advanced AI is no longer hypothetical.