Anthropic: Yemen Cell Tried AI Weapons
Anthropic says it blocked Claude accounts in Houthi-held Yemen that tried to develop advanced missiles with the AI model.
Anthropic has disclosed that users of its Claude chatbot in northern Yemen, territory controlled by Iran-backed Houthi rebels, attempted to use the AI model to develop advanced missiles — an episode that the company says ended in a failed test of a guided rocket rather than a working weapon.
According to Anthropic, the accounts behind the effort were blocked after the company identified them. The users, it says, did not succeed in "fielding an operational device," but they did carry out a test of a guided rocket that failed. Anthropic says it knows this because the users returned to Claude to ask why the test had failed.
The disclosure, made in a report released Thursday, points to a remote and rugged battlefield where AI is being applied to weapons development — an area already being reshaped by AI use from Ukraine to Gaza.
Three Weapons Programs in One Cell
Anthropic says it identified a cell in northern Yemen pursuing three different weapons programs. Those included a multi-variant missile designed to glide at hypersonic speed and a warhead that uses mobile phone hardware to maneuver mid-course.
Multi-variant missiles are ones in which different warheads and guidance systems can be fitted onto the same base design — a more economical way to build land, sea, and air weapons.
The company says the actors in Yemen used Claude Code rather than human software engineers to develop guidance, navigation, and control software. Before the accounts were banned, Anthropic says it had evidence the users had already built an offline simulation tool kit that does not use Claude or any other computing platforms.
Anthropic did not identify the users. But mountainous northern Yemen is controlled by the Houthis, suggesting the rebels are pursuing more sophisticated weapons while already wielding an array of drones, missiles, and other munitions in their campaign to seize more territory in Yemen and damage Saudi Arabia's oil exports.
The Houthi Response
Hazam al-Assad, a member of the Houthis' political bureau, rejected the idea that the group would depend on open sources to develop military capabilities.
"Our armed forces have modern, diverse and developed production capabilities and technology that it has accumulated over the period of Saudi aggression on Yemen," al-Assad said, referring to the kingdom's involvement in Yemen's 12-year civil war.
Al-Assad said all weapons are used for self-defense. He also described the suggestion that the Houthis would rely on open sources to develop and produce military capabilities as "unreasonable and illogical."
A Pattern of AI Misuse Reports
The report was the third Anthropic has put out since March 2025 on global misuse of its AI platforms. It described findings from December to August, ranging from state-sponsored groups spreading propaganda to unnamed actors researching how to make biological weapons more deadly.
Anthropic said it shared its findings with private and public partners. The company has previously reported on related misuse, including an account of Russian hackers using Claude to automate malware evasion.
The latest report arrives as AI's role in warfare continues to expand, with its use on a rugged and remote battlefield likely to increase concerns about its rapid spread.
Why Hypersonics Remain Out of Reach
Trevor Ball, a weapons analyst at Armament Research Services, said the Houthis "might be looking into hypersonic (missiles) by asking Claude," but they have nowhere near the production or technical capabilities to actually build them. He noted that U.S. hypersonic missiles "are still in testing."
Ball said the Houthis already possess Iranian-made anti-ship missiles with systems that allow the missiles to adjust guidance mid-course. He said the group is likely trying to develop its own capabilities to become less reliant on Iranian shipments of weapons and components.
Iran has boasted that it has hypersonic missiles and claimed it fired them at Israel last year. It denies arming the Houthis, which would violate a U.N. embargo, but Iranian weapons have been found on the battlefield and seized from shipments bound for Yemen.
Shipping Routes Under Pressure
The Houthis have launched waves of attacks on Saudi oil facilities and tankers in the Red Sea, threatening global trade as Iran continues to disrupt shipping in the Strait of Hormuz.
The rebels have made lightning territorial gains since Wednesday in their war with Yemen's Saudi-backed government, seizing strategic territories along the Red Sea coast and in the Bab al-Mandeb Strait, a key alternative to the Strait of Hormuz.
The group is known to have acquired and developed a sophisticated arsenal of weapons, including Iranian-made cruise and ballistic missiles with a range of more than 2,000 kilometers (1,200 miles) and unmanned submarines used in attacks.
Not 'Barefoot Tribal Fighters'
Adam Baron, a Yemen-focused researcher at the New America think tank in Washington, said the Houthis have kept pace with technological development.
"There's a tendency to see the Houthis as this group of barefoot tribal fighters, and that's just not true," he said. "Whether it's things like their emergent use of Claude, their use and manipulation of social media narratives or their ability to capitalize on the transfer of Iranian and wider axis expertise, we're talking about an incredible — and deepening — amount of institutional tech savvy."
Baron's assessment points to a group that has folded new tools into an existing technical base rather than starting from scratch.
The Numbers Behind the Arsenal
Several figures in Anthropic's account and the surrounding reporting help frame the scale of the activity and the weapons involved.
- Anthropic's report is the third it has published since March 2025 on global misuse of its AI platforms.
- The findings cover the period from December to August.
- The Houthis were pursuing three different weapons programs.
- The group's Iranian-made cruise and ballistic missiles have a range of more than 2,000 kilometers (1,200 miles).
- The rebels made territorial gains since Wednesday, seizing areas along the Red Sea coast and in the Bab al-Mandeb Strait.
- Saudi Arabia's involvement in Yemen's civil war has spanned 12 years.
What the Failed Test Leaves Open
Anthropic's account stops short of claiming that the Yemen-based users produced a working weapon. The company says they did not succeed in "fielding an operational device," and that the one test it refers to — of a guided rocket — failed. Its evidence for the failure is that the users turned back to Claude to troubleshoot.
The report does not name the users, and it does not specify how many accounts were blocked or which organizations, if any, were behind them. It places the activity in northern Yemen, territory the Houthis control.
The Houthi response, delivered by al-Assad, disputes the premise that the group needed outside help to develop military capabilities at all, describing such reliance as "unreasonable and illogical."
What remains undisputed in the reporting is the broader pattern: a well-armed non-state group operating in a strategically vital shipping corridor, and an AI developer documenting attempts to use its tools for weapons development alongside its other misuse findings.
Why It Matters
For security teams and policymakers, the episode suggests that the question is no longer whether AI will be applied to weapons research, but how quickly and by whom — and what safeguards can realistically catch it before an attempt succeeds rather than after a failed test sends users back to the chatbot for answers. Anthropic's ability to detect and block the accounts, and to describe what they were doing in enough detail to infer a failed rocket test, is the kind of visibility that vendors and regulators have been debating for years. The reliance on Claude Code for guidance, navigation, and control software hints at a workflow that could lower the barrier for groups without large engineering staffs. At the same time, the assessments from Ball and Baron caution against overstating what the attempt achieved: hypersonic production remains far beyond the Houthis' reach, and the group's existing arsenal already comes largely from Iranian transfers. For businesses that depend on Red Sea shipping, the more immediate concern remains the Houthis' territorial gains and attacks on tankers and oil facilities, which threaten trade regardless of how far their AI experiments progress.
Sources
- SecurityWeek Original source
Continue Reading
AI Rollouts Outpace M365 Permission Checks
A Syskit study finds 76% of UK and US organizations have deployed enterprise AI, but only 43% reviewed permissions first.
Anthropic Researcher Exits Over AI Safety Race
A researcher says he left Anthropic over concerns it and OpenAI prioritize competitive advantage over safety in AI development.
OpenAI Board Adds AI Safety Researcher
Paul Christiano, who pioneered a key training technique, joins OpenAI's foundation board and its safety committee, citing near-term loss-of-control risk.