SAN FRANCISCO, Oct. 11, 2026 | OpenAI says it banned a cluster of ChatGPT accounts operating from Iran that helped place about 100 articles under seven fake Western bylines in online outlets, according to a report the company published Oct. 8. OpenAI said the activity appeared consistent with a commercial actor running a for-hire influence campaign, but it could not identify the specific actor.
The articles appeared across roughly a dozen online outlets, mostly small to medium publications focused on international affairs, geopolitics and the Middle East, OpenAI said. The earliest identified article dates to July 2025 and the latest to October 2026, with output rising sharply after the US-Iran conflict began. OpenAI shared its findings with relevant authorities and industry partners.
How the Operation Worked
The operators wrote prompts in Persian and generated content in Persian and English, using VPNs to hide their location, per OpenAI. They used ChatGPT to review English drafts against outlets’ submission criteria, write pitch emails to editors, draft a bio for one persona, and produce batches of social media replies, including replies endorsing the operation’s own articles.
The seven personas each presented as a Western journalist: Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams and Alice Johnson. OpenAI said some had supporting social media accounts, and the Hoskins accounts showed Iran as their location despite a bio claiming he was “an American freelance writer.” Radio Free Europe/Radio Liberty describes Hoskins as having written dozens of articles critical of the US war with Iran, and reports the personas amplified one another’s content using AI-generated profile images. Many articles criticized the impact of US military action against Iran on Americans, according to Fox News’ coverage of the report. The Washington Post linked the Hoskins byline to articles in the Port St. Joe Star, Daily Kos and Middle East Monitor, Fox reported. Daily Kos and Middle East Monitor have since removed the articles.
How OpenAI Rated the Impact
OpenAI scored the article placements as Category 4 on its Breakout Scale, based on repeated breakout into media outlets, where six is the highest. The social media replies rated only Category 2, drawing low engagement, typically single or double digits. OpenAI also noted the operation’s own internal metrics were inflated because it counted views on the posts it replied to rather than on its replies.
Other platforms acted on the same personas: Meta banned an Ervin Hoskins Instagram account in August, and X suspended three persona accounts. OpenAI’s report counts 30 covert influence operations exposed since early 2024. RFE/RL quoted the company as saying AI can give such operations “greater scale, efficiency, linguistic fluency, and editorial ability,” and that operators can exploit editors and place content before audiences unaware of its origin. The finding is a reminder that content reaches readers through human editors, who are the point where these operations succeed or fail.
Frequently Asked Questions
What did OpenAI find about the Iran-linked ChatGPT accounts?
OpenAI said accounts operating from Iran used ChatGPT to help place about 100 articles under seven fake Western journalist personas across roughly a dozen online outlets. It banned the accounts and shared findings with authorities. It could not identify the specific actor.
How did the operation get articles published?
Operators used ChatGPT to edit drafts to outlets’ submission criteria and to write pitch emails to editors, then published under invented bylines. OpenAI says such operations can exploit editors who are unaware of the content’s origin.