Houthis used Claude AI to develop guided weapons, report says

A new report has raised fresh concerns about the growing misuse of artificial intelligence in military programmes after a group of threat actors in northern Yemen allegedly used Anthropic’s Claude AI to help develop software for guided weapons.

According to a threat intelligence report released by AI company Anthropic, the group was involved in three separate weapons-development programmes, including a guided rocket, a multi-stage ballistic missile and a missile concept incorporating a hypersonic glide vehicle. Anthropic said the actors used Claude Code to assist with guidance, navigation and control software.

The British Financial Times reported that the group appeared to be linked to Yemen’s Houthi movement, although Anthropic itself did not directly identify the actors as Houthis. The group was based in northern Yemen, an area largely controlled by the Iran-backed movement.

The disclosure provides an unusual glimpse into how advanced AI systems could potentially be incorporated into weapons-development efforts by non-state actors.

Anthropic said the group used Claude as a substitute for human software engineers while working on systems designed to steer and stabilise flying vehicles. The actors reportedly used multiple Claude sessions for different tasks, including software development, research and code review.

The company said the group also attempted to integrate an open-source autopilot with a phone-class flight computer and used AI assistance during simulation and testing.

One of the programmes involved a guided rocket that was actually test-fired in the field. Anthropic said the test appeared to have failed, after which the group returned to Claude to analyse what had gone wrong. However, the company said it did not have evidence that the actors successfully fielded an operational weapon.

The reported ballistic missile programme had a stated range of more than 2,000 kilometres, while another project involved a hypersonic glide vehicle. Anthropic said its visibility into the wider weapons programmes was limited, meaning it could not independently establish the full extent of their progress.

AI safeguards challenged

Anthropic said its safety systems blocked many of the group’s requests, but not all of them.

The actors reportedly attempted to evade safeguards by concealing their objectives, disguising the intended applications of the software and splitting their work across multiple conversations. Anthropic subsequently banned accounts associated with the operation and shared relevant information with public and private-sector partners.

The company said the case demonstrates a broader challenge facing the AI industry: increasingly capable models can provide useful assistance not only to legitimate engineers and researchers but also to actors seeking to develop weapons.

Anthropic’s latest report identified several cases involving conventional weapons, including missiles, armed drones and other military systems. The company said it has since introduced additional classifiers designed to detect and block requests related to weapons development.

The revelations have added to growing international debate over whether AI companies can keep increasingly powerful systems from being exploited for military and other dangerous purposes.

For now, Anthropic’s findings do not establish that AI enabled the Houthis to successfully produce an operational advanced missile. But the reported test of an AI-assisted guided rocket offers a striking warning of how quickly sophisticated technology can move from the digital world into real-world conflict.