The Claude app, made by AI company Anthropic, is shown on a smartphone. Anthropic said it banned an Iran-linked account that used Claude to compile targeting handbooks on U.S. naval forces, including personnel names pulled from public military photo captions and satellite-imagery query scripts. (Anthropic)
NAPLES, Italy — A major U.S. artificial intelligence developer recently disrupted an Iran-connected effort to use the technology to gather and analyze public information to aide in targeting U.S. naval forces in the Middle East.
San Francisco-based Anthropic discovered, investigated and banned an unidentified user it linked to Iran from accessing its AI assistant Claude, the company said in a report released Thursday detailing its efforts to root out misuse of AI.
The user exploited Claude to compile a roster of U.S. personnel scraped from captions on public military photographs, collect publicly accessible ship and aircraft transponder identifiers and commercial satellite-imagery query scripts and generate an inventory of public websites that exposed U.S. naval movements, according to the report.
Anthropic did not say when it discovered the misuse of Claude for targeting U.S. forces. The report covered incidents discovered and disrupted from December to August.
“The threat actor also directed Claude to compile vulnerability research on shipboard systems,” the report stated.
The effort included using AI to develop “targeting books” to identify and track Navy positions based on open-source information, according to the 154-page report, which detailed examples of Anthropic’s efforts to expose misuse of the technology in cyber, surveillance and influence operations.
The report also described a China-based actor’s use of Claude to prepare a briefing on U.S. anti-torpedo weapon systems using open-source reporting.
In addition, it detailed efforts to curb the use of AI in scams and fraud, conventional and biological weapons development, and the theft of proprietary AI capabilities for use in another model.
The cases shared weren’t typical misuse but rather examples of the most notable and novel threat activity identified to date, Anthropic said in the report, issued because of what it described as a sense of responsibility to disclose malicious misuse.
“As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer,” Anthropic said.
The Anthropic disclosures come shortly after the Navy warned its personnel about security risks.
A sailor stands missile systems supervisor watch aboard the destroyer USS Gonzalez, April 25, 2026. Anthropic said an Iran-linked account used its Claude AI to compile targeting recommendations against U.S. naval forces, including research into vulnerabilities in shipboard communications and control systems. (Mark Peña/U.S. Navy)
Service members, civilians and families were urged last month to take precautions against potential threats by doing such things as removing details from social media that could reveal their Navy affiliation.
The guidance also advised personnel to report any sightings of drones near bases, people lingering near or photographing installations, or efforts to obtain information about ships, personnel or deployment dates, among other things.
“By embracing this heightened posture of vigilance, we will safeguard our force, protect our families, and ensure the unhindered execution of our national security mission,” acting Navy Secretary Hung Cao said in an Aug. 19 message to the force.
The message followed recent events of “uncrewed aircraft systems encounters, suspicious activity, cyber threats and acts of violence,” Cao wrote.
