Key Points
- Yemen-based cell sought AI help for guided rockets and missiles with 2000 km range goal
- Anthropic confirmed test-firing of guided rocket that failed, prompting return to Claude
- Houthis claimed control of Red Sea island on Friday completing Bab al-Mandab push
A cell based in northern Yemen attempted to use Anthropic’s Claude chatbot to develop software for guided rockets and missiles, the US artificial intelligence company disclosed in a report published on Thursday (10 September 2026), raising fresh concerns about the misuse of generative AI for weapons development.
Anthropic said it had identified three separate weapons programmes run by the Yemen-based group. These included a guided rocket using a commodity phone-class flight computer, a basic processor similar to those found in smartphones, with final-phase homing guidance that would allow the projectile to steer towards a target in its last moments of flight.
By the numbers
Key figures from this story- 2,000 km
- Stated range goal for ballistic missile programme
- 3
- Weapons programmes identified by Anthropic
The second programme involved a multi-stage ballistic missile, a weapon designed to travel outside the atmosphere before re-entering to strike a distant target, with a stated range goal above 2,000 kilometres. The third was a multi-variant missile that included a hypersonic glide vehicle variant, a type of weapon that travels at more than five times the speed of sound and can manoeuvre to evade defences.
According to Anthropic, the cell used Claude to develop guidance, navigation and control software for these airborne weapons. The company said this work would have otherwise required a team of engineers with specialised expertise.
Rocket test failure
Anthropic said it did not have evidence that the actors succeeded in fielding an operational device. However, the group did test-fire a guided rocket.
“This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed,” Anthropic stated in its report.
The company said it had banned the relevant accounts as soon as it had detected the illegal activity. The report did not name the actors involved, but the location in northern Yemen points to the Houthis, the Iran-backed movement that controls much of the region including the capital Sanaa.
The disclosure comes as the Houthis have launched a wide-ranging offensive across Yemen towards the strategic Bab al-Mandab strait, a narrow waterway connecting the Red Sea to the Gulf of Aden that serves as a critical shipping lane between Europe and Asia.
On Friday (11 September 2026), the Houthis claimed control of a Red Sea island in the middle of the Bab al-Mandab shipping lane, completing their push to dominate the area. The claim, if verified, would give the group significant leverage over international maritime traffic.
The Anthropic report forms part of a broader disclosure by the company on how it had prevented illegal uses of Claude across several countries. The findings underscore growing concerns among governments and technology companies about the potential for generative AI systems to lower the barrier for weapons development by non-state actors.
AI safety researchers have warned that large language models could provide guidance on tasks that previously required scarce specialist knowledge, from synthesising chemical compounds to designing weapons systems.
Anthropic, founded in 2021 by former OpenAI researchers, has positioned itself as a safety-focused AI company and publishes regular reports on misuse attempts it detects on its platform.
The company did not disclose how long the Yemen-based cell had been using Claude before detection, or what specific safeguards failed to prevent the initial access.
Your Questions, Answered
What weapons did the Yemen cell try to develop using Claude AI?
According to Anthropic, the cell worked on three programmes: a guided rocket with homing guidance, a multi-stage ballistic missile with a 2,000 km range goal, and a multi-variant missile including a hypersonic glide vehicle.
Did the Yemen group successfully build a working weapon?
Anthropic said it has no evidence the actors fielded an operational device. However, they did test-fire a guided rocket that failed, prompting them to return to Claude to diagnose the failure.
Who are the actors behind the weapons programmes?
Anthropic did not name the actors, but the location in northern Yemen points to the Houthis, the Iran-backed movement controlling much of the region including the capital Sanaa.
What action did Anthropic take after detecting the misuse?
Anthropic said it banned the relevant accounts as soon as it detected the illegal activity and published details in its report on preventing misuse of Claude.
