Anthropic said Islamic Republic government agents used the AI model “Claude” to review social-media content inside Iran and identify potential opponents and sensitive political content.

According to a new report by Anthropic, users in countries including Iran, China, and Russia have bypassed geographic restrictions for access to Claude by using private virtual networks, fake or stolen accounts, and intermediary services.

The company also reported identifying nine related intrusion operations involving countries including Iran, Russia, and Turkey, in which government and private actors played a role. Some of these operations targeted audiences on six continents, and the timing of several of them coincided with elections in Moldova and Kenya.

Anthropic added that more of the AI-generated content in these operations received little real engagement or had no engagement at all before the activities were halted. The company said it disrupted misuse and provided relevant information to governments and relevant institutions.