Anthropic — the U.S. AI company behind Claude — has published a threat-intelligence report describing a coordinated influence operation linked by its investigators to the Mujahedin-e Khalq (MEK/PMOI) and the National Council of Resistance of Iran (NCRI).
What makes the report disturbing is not simply the use of AI. It is the deception.
According to Anthropic, at least four operators involved worked for official NCRI media. In one case, about 8,400 Telegram posts from a real political activist were fed into an AI system so it could imitate his language and style. The system was then used to assist conversations with people in his real network who, according to Anthropic, did not know they were interacting through an AI-assisted impersonation.
The operation went further. More than 500 social-media channels were scraped, and people were categorized by factors including location, age, profession, political orientation and previous arrests. Around 51,944 archived messages were analyzed to build political and psychological profiles of specific individuals in Iran.
Anthropic also describes AI-generated personas made to look like ordinary Iranians while promoting Maryam Rajavi’s political program, coordinated Instagram accounts, and efforts to conceal the MEK connection so political messaging could appear independent or grassroots.
This is not normal political persuasion. It is manufactured authenticity: creating the appearance of spontaneous public opinion while hiding the organized actor behind it.
That matters because people form political judgments partly by observing what they believe other people think and support. If coordinated accounts, artificial identities and fabricated content create a false impression of popularity or consensus, the information environment itself is being manipulated.
The legal implications may also be serious. Political opinions are treated as sensitive personal data under European data-protection law and Albanian law. Systematically collecting political information about identifiable people, profiling them and using those profiles for targeted influence may therefore raise major questions about privacy, lawful processing and transparency. Identity impersonation can also be unlawful under national laws, depending on the jurisdiction and harm caused.
This does not by itself prove that the MEK, NCRI or any named individual committed a specific crime. Anthropic is not a court, and criminal liability depends on evidence, jurisdiction and who actually controlled the operation.
There is also an important limitation: while Anthropic identified operators connected to official NCRI media, common methods and references to MEK leadership, it says it could not determine the precise degree of centralized control by the MEK’s senior leadership.
That distinction matters. Criticism should be evidence-based.
But the democratic question remains: a movement claiming to offer Iran a democratic future must also be judged by the methods it uses to gain influence today.
If support is genuine, it should not need to be simulated.
If legitimacy is genuine, it should not need to be manufactured.
Democracy cannot be built by fabricating the very public opinion later presented as proof of democratic legitimacy.
Amir Yaghmai