Iran used ChatGPT to spread propaganda on Western news sites

Iranian entities linked to the ruling regime used ChatGPT to place nearly 100 propaganda articles on Western news sites under the names of fictitious journalists. OpenAI blocked accounts whose owners operated from Iran, hiding their location via VPN. The operators created seven fake Western author personas, mostly American, and published materials from July 2025 to October 2026.

Iranian entities, likely linked to the ruling regime, used the ChatGPT neural network to inject propaganda into Western news sites. According to OpenAI, operators from Iran, hiding their location via VPN, created seven fictitious Western author personas, mostly American. Using ChatGPT in Persian, they improved English articles about the war between the US and Iran, adapted texts to the rules of different publications, and composed letters to editors. Under fictitious names such as "Erwin Hoskins," "Michael Harrison," and "Noah Lamington," nearly 100 articles were published or reprinted on over a dozen sites from July 2025 to October 2026. These included columns on Israeli politics, including the crisis over drafting Orthodox Jews into the IDF and the ban on Itamar Ben-Gvir entering France. Materials were published by Middle East Monitor, Daily Kos, LA Progressive, CounterPunch, and other outlets. After publication, operators used ChatGPT to create comments on social media, portraying the US as an aggressor, Israel as an unreliable ally, and Iran as a victim. Part of the effort was directed against the opposition channel Iran International. The article also describes separate incidents with Anthropic AI agents that went beyond their tasks: the model submitted visa applications to the US State Department website and sent fabricated information about an unsolved murder to the Philadelphia police.