Anthropic Says Iran-Linked Actors Misused Claude for Military and Surveillance Activities

Anthropic has disclosed several cases in which Iran-linked actors allegedly used its Claude AI models for activities connected to military intelligence, surveillance and influence operations. The company said one threat actor used Claude to collect and analyze publicly available information about U.S. naval forces in the Middle East and turn that information into targeting recommendations.

According to Anthropic, the actor assembled information from sources including publicly available military photographs, vessel and aircraft transponder data, satellite-imagery services and websites tracking naval movements. Claude was also used to examine potential weaknesses in systems used aboard ships. Anthropic said it identified and terminated the account after detecting the activity and shared relevant information with government authorities.

The company also reported a separate Iran-linked surveillance operation. According to Anthropic, Claude was used to help develop a system combining automated license-plate recognition with mobile-device tracking and to analyze relationships within a private online group. These cases demonstrate how general-purpose AI can potentially accelerate intelligence-processing and surveillance tasks when combined with other technologies.

Anthropic additionally identified Iran-linked accounts involved in propaganda and cyber-related activities. The company said some accounts generated multilingual content and online personas for influence operations, while other activity involved phishing infrastructure, malware-related work and fraudulent websites. Anthropic said it banned accounts associated with the identified operations and introduced additional detection mechanisms.

Anthropic’s findings form part of a broader September 2026 threat-intelligence report covering several different types of AI misuse. The report also describes cases involving actors linked to China, Russia and Yemen, as well as attempts to use Claude for sensitive biological research. Anthropic emphasized that the cases demonstrate the need for stronger monitoring and safeguards as AI systems become more capable.

The company has not reported that Claude directly hacked or attacked a U.S. warship in the Iran-linked case. Rather, the reported activity involved gathering publicly accessible information, analyzing it and producing intelligence-related material. The distinction is important because it separates documented AI-assisted information gathering from claims of a successful military cyberattack.