Anthropic said Friday that a group linked to the Iranian government misused its Claude AI model to help gather intelligence for potential attacks on U.S. Navy warships, marking one of the most serious documented cases of a state actor weaponizing commercial AI against American military assets. The San Francisco-based AI company disclosed the incident in a threat intelligence report published this week, saying it detected the activity, banned the associated accounts, and notified U.S. authorities.
According to Anthropic, the operatives used Claude to research vessel specifications, radar and defense systems, and vulnerabilities in U.S. naval assets operating in the Middle East. The company said the requests were disguised as legitimate research queries, a technique known as prompt obfuscation, designed to slip past the model’s safety filters.
What Anthropic Found
In its report, Anthropic said its internal threat-detection systems flagged unusual patterns of activity from accounts traced back to Iran, including attempts to extract technical details about ship classes, propulsion systems, and countermeasure technology. The company said it could not confirm whether the information was used in any actual operation against U.S. forces, but characterized the activity as consistent with reconnaissance for a potential strike.
“We identified a cluster of accounts, assessed with high confidence to be linked to Iranian state interests, attempting to use our models to support hostile military planning against U.S. naval assets,” Anthropic said in its report. “We terminated the accounts, reported relevant details to U.S. government partners, and are strengthening safeguards against similar attempts.”
The company did not name a specific unit within Iran’s government, but said the pattern of behavior — including the use of intermediary accounts and layered prompts — matched tactics previously associated with Iranian state-linked cyber units.
Context: Rising Tensions in the Gulf
The disclosure comes amid heightened friction between Washington and Tehran, with U.S. Navy vessels maintaining an elevated presence in the Persian Gulf and Strait of Hormuz throughout 2026. Oil markets have already reacted sharply to the broader conflict, as detailed in NarwhalTV’s earlier coverage of oil prices surging past $100 a barrel amid the Iran war escalation. Analysts say any confirmed attempt to target U.S. military assets, even one intercepted before causing harm, is likely to further rattle markets and diplomatic relations.
The Pentagon has not issued a formal statement confirming or denying operational impact from the alleged targeting attempt. A defense official, speaking on background, said the military “takes any reported threat to naval assets seriously” and is reviewing the Anthropic report in coordination with intelligence agencies.
A Broader Pattern of State Misuse
Anthropic’s disclosure is the latest in a string of reports from major AI developers documenting attempts by foreign governments to exploit commercial AI tools for military, intelligence, or influence operations. Earlier this year, the company also faced scrutiny over its own data practices, following reports covered by NarwhalTV on claims that Anthropic tracks AI critics, which the company denied.
Security researchers say the Iran case illustrates a growing challenge for AI developers: distinguishing between legitimate research queries and disguised attempts to extract sensitive military or technical information. Anthropic said it has since tightened detection systems for queries involving weapons systems, military hardware, and defense infrastructure, and is working with U.S. national security agencies to share indicators of similar misuse.
Iran’s Response
Iranian officials have not publicly commented on Anthropic’s specific allegations. Tehran has previously denied similar accusations of using foreign AI tools for military targeting, characterizing such reports as part of a broader U.S. campaign to justify sanctions and military posturing in the region.
What Happens Next
Members of Congress from both parties have signaled interest in reviewing the incident, with some lawmakers calling for hearings on how AI companies detect and report foreign state misuse of their platforms. The episode is also expected to feature in ongoing debates over federal AI safety legislation, following recent polling showing strong public support for stricter oversight of advanced AI systems.
For now, Anthropic says it has no evidence that the targeting attempt resulted in any successful operation against U.S. forces, and emphasized that its detection systems worked as intended in flagging and halting the activity. But the case underscores a broader reality facing AI developers: as their models become more capable, they are also becoming targets — and tools — in state-level conflicts far beyond the tech industry’s traditional concerns.
NarwhalTV will continue to follow developments in this story, including any official U.S. government response or further disclosures from Anthropic.
