Anthropic says an Iran-linked group used its Claude AI models to compile intelligence and targeting material aimed at US Navy warships in the Middle East. The finding appears in a 154-page September 2026 threat report titled “Detecting and Countering Misuse of AI,” published by the AI company.

According to Anthropic, the group used Claude to build an automated system that collected public military photographs, commercial satellite imagery and data from public websites tracking naval movements. It also extracted the names of US personnel from public photo captions and researched vulnerabilities in shipboard communication equipment. Anthropic says the account was also linked to an Iranian domestic surveillance project.

The company said it could not confirm whether the compiled material was later used to plan or support any operation against US forces. It banned the account, added new detection measures and shared its findings with government authorities.

The same report describes a Houthi-linked cell in Yemen using Claude Code to help develop three weapons programs, including a ballistic missile with a range over 2,000 km, and says its models were also exploited in cyber and influence operations tied to Russia and China.