For decades, Iran’s clerical dictatorship has treated propaganda as an instrument of state power. Newspapers, television, mosques, cultural centers, intelligence agencies, and diplomatic missions have all played their part in spreading the regime’s message and demonizing its enemies. Now the mullahs have acquired a powerful new weapon. Artificial intelligence.
A disturbing investigation published by Anthropic has exposed three Iranian state-aligned influence operations using its Claude artificial intelligence system to construct sophisticated propaganda campaigns. These were no collection of enthusiastic amateurs posting slogans from Tehran. Anthropic says the operations were linked to Iranian state propaganda institutions and formed part of what their organizers themselves described as a “soft war” and “cognitive warfare” program.
The implications are chilling. According to Anthropic, Claude became what amounted to an administrative and operational headquarters. It helped create campaign plans, operational manuals, artificial personas, target databases, amplification schedules, and ministerial planning documents. Government intelligence bulletins could be transformed rapidly into material in Farsi, Arabic, Urdu, Malay, Spanish, and English, with ambitions extending to 20 languages. Even more sinister was the deliberate laundering of attribution. State propaganda could be rewritten to appear as the work of foreign journalists, independent commentators, or ordinary citizens. Artificial personas could manufacture the illusion of spontaneous public opinion. This represents industrialized deception.
The objective is familiar to anyone who has studied the Islamic Republic. Tehran has always understood that propaganda becomes more effective when the audience cannot see who is pulling the strings. One Iranian official reportedly explained the philosophy perfectly. Cultural attachés should act as directors rather than narrators. In other words, the regime wants others apparently to speak for it. Artificial intelligence makes that infinitely easier.
The latest revelations form part of a much wider pattern. In August 2024, OpenAI closed accounts connected with an Iranian influence operation known as STORM 2035. The network used ChatGPT to generate articles and social media material dealing with the American presidential election and other politically divisive subjects. OpenAI subsequently found renewed activity from the same operation generating English and Spanish posts through accounts posing as residents of the United States, Britain, Ireland, and Venezuela. Another OpenAI investigation uncovered links between AI-generated material and the International Union of Virtual Media, an Iranian influence network. Content appeared through supposedly independent websites and social media accounts, obscuring its Iranian origins.
Microsoft has documented an equally disturbing record. During the 2024 American election, Iranian actors created covert news sites targeting voters on opposing sides of the political spectrum. The purpose was less about persuading everyone toward one position than aggravating existing divisions and undermining trust. Microsoft also identified Iranian cyber operations combining stolen information, fabricated identities, propaganda, and covert social media activity.
Following the Hamas attack on Israel in October 2023, the technique became still more aggressive. Microsoft recorded Iranian influence operations combining cyber-attacks with social media manipulation. In December 2023, streaming television services were interrupted by a fabricated news broadcast featuring an apparently AI-generated presenter. Audiences in Britain, Canada, and the United Arab Emirates were among those reached. The danger therefore extends far beyond Iran. AI allows a dictatorship with limited resources to manufacture content on an extraordinary scale. One operator can create hundreds of identities. One propaganda unit can communicate in numerous languages. One intelligence bulletin can become thousands of posts, articles, videos, and comments. Falsehoods can be repeated until they appear credible simply because people encounter them everywhere.
For the Iranian regime, this technology fits perfectly with an established strategy of demonization. Opposition figures can be smeared. Dissidents can be portrayed as terrorists or foreign agents. Religious minorities can be targeted. Western politicians can be attacked or manipulated. Fabricated stories can be attributed to respected institutions. Anthropic discovered precisely this technique. One Iranian network falsely attributed claims to respected Western organizations, including CSIS, Brookings, and RAND. The objective was obvious. Borrow somebody else’s credibility to sell Tehran’s propaganda.
The West must respond with urgency. First, artificial intelligence companies should deepen cooperation with democratic governments, security agencies, universities, and one another. When one company discovers a state influence network, indicators should rapidly be shared across the industry. Second, social media companies must become far more effective at identifying coordinated artificial personas and networks. Removing a single account achieves little when hundreds more can appear within hours. Third, governments should establish specialist units capable of publicly exposing foreign influence campaigns quickly. Exposure is itself a weapon. Propaganda loses much of its power once the audience knows Tehran is directing it.
Fourth, media organizations, parliamentarians, and think tanks must authenticate explosive claims before repeating them. A quotation carrying the name of RAND, Brookings, CSIS, Reuters, or any respected institution should never acquire credibility merely because an impressive graphic accompanies it. Fifth, Western governments should impose sanctions on Iranian officials and institutions directing covert foreign influence operations. Propaganda warfare conducted by state agencies deserves consequences. Finally, democratic societies must defend the principle of attribution. People have every right to advocate Tehran’s policies if they choose. They equally have a right to know when supposedly independent voices are manufactured by a dictatorship.
Artificial intelligence did not invent the mullahs’ machinery of lies. It has given that machinery speed, reach, and sophistication unimaginable a decade ago. The battle against Iran’s clerical regime is therefore being fought on another front. Tehran calls it cognitive warfare. The democracies should start treating it as exactly that.
Struan Stevenson was a member of the European Parliament representing Scotland (1999–2014), president of the Parliament’s Delegation for Relations with Iraq (2009–14), and chairman of the Friends of a Free Iran Intergroup (2004–14). He is an author and international lecturer on the Middle East.