Iran‑linked group used Claude AI to track U.S. Navy ships, Anthropic says
Anthropic says an Iran‑linked group used its Claude AI system to combine ship and aircraft transponder data, military photos, and commercial satellite images in an attempt to identify U.S. Navy vessel locations and potential communications vulnerabilities in the Middle East.
An Iran‑linked group turned to a commercial AI system to help follow U.S. warships in the Middle East, according to the system’s developer. The case shows how quickly off‑the‑shelf artificial intelligence is being pulled into state‑aligned intelligence work.
Anthropic told the Wall Street Journal that operators associated with Iran used its Claude model to assist in tracking U.S. Navy vessels. The group wasn’t hacking into classified networks. Instead, it pulled together public sources: ship and aircraft transponder data, military photographs, and commercial satellite imagery. Claude was then asked to help identify ship locations and potential communications vulnerabilities.
In practical terms, the model was being used as an analyst that could process multiple open data feeds and point the users toward useful patterns. According to Anthropic’s account, the company detected the activity, disrupted the operation, and banned the accounts involved.
The same reporting describes a broader pattern of misuse attempts. One Yemen‑based group, for example, used multiple instances of a Claude model optimized for coding while working on guided rockets, ballistic missiles, and what was described as a hypersonic‑glide project. Even when the system is designed to refuse direct help on building weapons, persistent users can try to break those safeguards into smaller technical requests.
For U.S. naval forces, the risk isn’t that AI suddenly reveals hidden secrets. It’s that it makes it far easier and faster to turn open information into something closer to targeting data. Ship transponder signals, aircraft tracks, and overhead imagery have long been available in some form. An AI model that can fuse them at scale changes how quickly an adversary can understand deployments and communications habits.
The episode underlines how general‑purpose AI tools blur the line between civilian and military technology. With nothing more than an account and an internet connection, operators can attempt tasks that once demanded dedicated software and specialist teams.
Signals to watch now include whether U.S. authorities publicly attribute the activity to specific Iranian units, whether militaries tighten rules around public transponders and images near sensitive deployments, and how aggressively AI companies move to monitor and block high‑risk national‑security use cases on their platforms.
Sources
- OSINT