A weapons engineering cell in Yemen used Anthropic’s Claude artificial intelligence system to help develop missile guidance software - and turned back to the AI tool within hours of an apparently failed guided-rocket test to work out what had gone wrong, according to a new threat intelligence report from the US technology company.The cell used Claude Code while working on three weapons programmes, including a multi-stage ballistic missile with a stated range goal of more than 2,000km and a missile family referred to as the “R2000” that included a hypersonic glide vehicle variant, according to Anthropic’s findings.AI used for weapons developmentThe Yemen case was among six conventional-weapons cases identified by Anthropic.Reuters reported that Anthropic had uncovered actors using Claude to develop software for conventional weapons including missiles, armed drones, firearms, bombs and other munitions, as well as the targeting and control systems used to operate them.According to the Times of India’s account of Anthropic’s findings, the Yemen-based cell was working on three programmes: a guided rocket, a multi-stage ballistic missile and the R2000 missile family.The company said the group used Claude Code for engineering work involving guidance, navigation and control software.Anthropic’s findings suggest the actors were using the AI system for more than simply obtaining technical information. The company said Claude was being used to carry out software engineering tasks associated with the weapons projects.The specifications and capabilities described in the report were based on activity observed by Anthropic and have not been independently verified.Back to Claude after failed testOne of the most striking episodes described in the report involved an apparent test of the guided rocket.According to the Times of India, citing Anthropic’s findings, the test appeared to fail. Within hours, members of the cell returned to Claude and used the AI system to help investigate what had gone wrong.Anthropic did not establish that the group had succeeded in fielding an operational weapon.The company also did not identify the Yemen-based cell as belonging to a specific organisation.Six weapons-related casesThe case forms part of a much broader investigation by Anthropic into potentially harmful uses of its AI systems.Reuters reported that the company had identified what it called “new categories of threat actors” using Claude to develop weapons-related software or support intelligence gathering and procurement connected to weapons programmes.The company said the cases demonstrated how increasingly capable AI systems could be used for specialised technical work associated with weapons development.Anthropic has also been investigating the use of its systems in cyber operations.Growing capabilities raise concernJacob Klein, Anthropic’s head of threat intelligence, told Reuters that AI models had become significantly more capable over the past year.“A year ago, let’s say you wanted to optimise a drone or optimise the software on a missile, the models just wouldn’t be as good at that task as they are now,” Klein said.Anthropic said cybercriminals and state-backed actors were increasingly using AI not merely to answer questions or assist with individual tasks, but to carry out larger and more complex portions of operations.The company said it had detected and disrupted the activities outlined in its report and was using what it learned from such cases to strengthen its safeguards.- With inputs from Reuters