A group linked to Yemen reportedly used Anthropic’s Claude AI while working on several weapons-related programs.
The findings were detailed in a threat intelligence report released by Anthropic.
According to the company, Ansar Allah used Claude while developing software related to guidance, navigation and control systems.
Anthropic said the activity involved a guided rocket and missile-related projects.
Anthropic Identifies Three Weapons Programs
Anthropic said it identified a cell associated with Ansar Allah that was working on three separate weapons programs.
One project involved a guided rocket, according to the report. Another involved a multi-stage ballistic missile with a stated range target exceeding 2,000 kilometers.
The third was described as a multi-variant missile program.
The report indicates that artificial intelligence was used to support some software development activities linked to these projects.
Anthropic said the group used Claude Code during its work.
The reported activity highlights growing concerns about advanced AI tools being misused for military purposes.
Claude Used for Flight Software Development
According to Anthropic, the group used its AI system while developing software for controlling and stabilizing flying vehicles.
The company said the activity included work involving an open-source autopilot and a phone-class flight computer.
Claude was also reportedly used in connection with flight simulation work. Anthropic further reported that the group conducted a test launch of a guided rocket.
The company did not suggest that Claude independently designed or manufactured the weapons.
Instead, the report describes AI being used as a technical assistance tool during software-related development.
The reported case involving Yemen demonstrates how general-purpose AI systems can attract misuse attempts from sophisticated actors.
Anthropic Says Safeguards Blocked Requests
Anthropic said its safety systems prevented many requests associated with the activity. However, the company reported that users attempted to bypass those safeguards.
According to the report, some requests concealed the broader purpose behind the software work.
Tasks were also reportedly divided across different sessions. This approach made it more difficult to identify the full objective from individual interactions.
Anthropic said it eventually identified the activity and took action against the accounts involved.
Accounts Banned Following Investigation
The company said the relevant accounts were banned after the activity was detected. Anthropic also shared information about the case with public and private sector partners.
The incident adds to wider concerns about preventing powerful AI systems from assisting weapons development.
AI companies have increasingly introduced safeguards designed to detect potentially dangerous requests.
However, threat actors may continue attempting to disguise their objectives or divide complex tasks into smaller requests.
The reported Yemen case shows the challenges AI companies face when monitoring sophisticated misuse.
Anthropic’s findings are the company’s assessment of activity observed on its systems. The report does not mean that the AI platform independently created the weapons described.
In other news read more about: Modi Government Faces Fresh Economic Questions as BRICS Summit Puts India in Spotlight
Instead, Anthropic says its technology was used during parts of the software development process before the accounts were banned.
