An Iran-linked threat actor used Anthropic's Claude AI platform to compile targeting handbooks against U.S. Navy forces in the Middle East, according to a new company report.
An Iran-linked threat actor used Anthropic's Claude artificial intelligence platform to assemble targeting material on U.S. Navy forces in the Middle East, the company disclosed in a threat report released Thursday. The activity spanned December 2025 through August 2026 and stands as one of the clearest public examples yet of a hostile regime converting a commercially available AI model into a digital reconnaissance tool. Anthropic said it shut the account down, built new safeguards, and shared what it found with the federal agencies responsible for countering the threat.
The disclosure lands as American naval power remains heavily engaged across U.S. Central Command's area of responsibility, where carrier strike groups and guided-missile destroyers continue to deter Iranian aggression and protect the Strait of Hormuz. For the sailors standing those watches, the report is a reminder that the battlefield now includes the open internet -- and the AI systems that can sort through it in seconds.
The threat actor used Claude to compile what the report describes as "targeting handbooks" designed to identify and track naval positions using open-source information. None of the raw material was stolen. It was already public.
The account pulled names and details on U.S. service members from publicly released military photographs, queried ship and aircraft transponder data that broadcasts identifying codes and locations, gathered commercial satellite-imagery scripts used to pinpoint exact geography, and catalogued public websites that track Navy movements. Individually, each source looks mundane. Combined and organized by a large language model, they form a picture that hostile intelligence services once needed entire teams to assemble.
Anthropic also said the actor used Claude to probe potential weaknesses in U.S. Navy shipboard systems -- research that could feed cyber intrusions, electronic warfare, or physical attacks. The company did not say whether the compiled intelligence was ever used in an actual Iranian operation.
The Iran case was not isolated. The same report describes a cell of Yemeni operatives that used Claude to develop targeting software for guided missiles. Anthropic said its investigators found that the group test-fired a guided rocket, though the company said it has no evidence the actors succeeded in fielding an operational weapon.
Taken together, the findings show how commercial AI is lowering the technical barrier for regional proxies and terror networks. What once required specialized missile engineers or state-level signals intelligence can now be approximated by anyone with a subscription and a novel way to phrase a request.
Anthropic said it banned the offending accounts and added safeguards meant to block similar misuse. The company framed its 154-page report as part of an effort to be transparent about how its models are being abused -- and to help governments respond faster.
The report sharpens a debate Washington has been having for more than a year: how to keep advanced American AI in the hands of U.S. warfighters while denying it to adversaries. The Pentagon has pushed hard to expand military use of frontier models, and the Trump administration has made American AI dominance a central plank of its national security agenda.
Earlier this year the U.S. military ended its relationship with Anthropic after the company refused to accept Pentagon terms that it said could open the door to mass surveillance or fully autonomous weapons. That standoff cut American commanders off from a tool that adversaries, by all indications, were never going to hesitate to exploit.
The contrast is stark. Tehran's operatives did not pause for ethics reviews, corporate policy debates, or legal carve-outs. They simply pointed a chatbot at open-source data and started building target folders.
The most immediate question is whether the intelligence Anthropic uncovered will translate into hardened operational security for deployed forces. Force protection now has to account for what a foreign analyst can learn from a photograph caption, a public flight-tracking site, or a satellite image order.
Lawmakers are likely to press for clearer rules on how AI companies detect and report hostile foreign activity. Expect new scrutiny of whether the government gets timely warnings when an adversary turns an American model into a weapon -- and whether Washington can close the gap between the private labs building these systems and the commanders who depend on them.
Anthropic says it has already begun building those bridges. Whether that is enough will be measured against a threat that is only getting faster.
An Iran-linked threat actor used Anthropic's Claude AI platform to compile targeting handbooks against U.S. Navy forces in the Middle East, according to a new company report.
freedomlinedispatch.com
