OpenAI said Thursday it shut down an Iran-linked influence campaign that used ChatGPT to polish articles, tailor submission emails and manufacture social-media reactions, helping place or syndicate nearly 100 pieces about the U.S.–Iran war in real news outlets under fake bylines. The disclosure lands as governments and publishers already argue over how generative tools lower the cost of covert propaganda.
What investigators tied to the network
In a threat report released Thursday, OpenAI described the Iranian activity under the label “Bogus Bylines.” The company banned a cluster of ChatGPT accounts it traced to Iran that supported seven personas posing as Western journalists, mostly Americans writing on international affairs and the Middle East.
OpenAI said it identified almost 100 articles published or syndicated across roughly a dozen online outlets between July 2025 and October 2026, with output accelerating after this year’s widening conflict. The earliest article in the set dated to July 2025; the latest appeared this month. Titles cited in accompanying reporting included pieces questioning U.S. policy toward Iran and framing Western credibility during the war.
The firm rated the article-placement portion of the campaign at Category 4 on its six-point scale for how far influence operations break beyond their own networks—higher than typical bot-driven spam because the stories reached third-party publishers.
How ChatGPT was used in the workflow
According to OpenAI, operators prompted the model in Persian and English to edit drafts, adapt copy to individual outlets’ submission rules and draft pitch emails to editors. After articles went live, they asked ChatGPT to generate batches of comments portraying readers reacting to the stories, including narratives that cast the United States as the aggressor and Iran as a resilient target of escalation.
OpenAI said the users also requested English and Persian comment sets tied to missile strikes, ceasefire talk and diplomacy. In at least one case, transparency data linked a promoted account to a West Asia–based mobile application, and Instagram metadata pointed to Iran, screenshots in the report showed.
The company stressed it could not name the ultimate sponsor. “The overall activity appeared consistent with a commercial actor running a for-hire influence campaign, but we are unable to identify the particular actor involved,” the report said.
Parallel Russian operation in the same release
Thursday’s release paired the Iranian case with a separate Russia-linked effort OpenAI said involved a fabricated think tank that hired unwitting local staff to produce anti-Ukraine material. That campaign received a higher severity score and, according to OpenAI, was the first in its tracking to draw public comment from politicians in multiple countries. The two cases share a pattern: state-aligned actors treating large language models as editing rooms rather than single-click fake-news factories.
What publishers and regulators face next
Security researchers have warned for two years that AI-assisted pitching could slip past overstretched editorial desks, especially at smaller geopolitics sites. OpenAI’s ban blocks the identified accounts but does not remove stories already indexed by search engines or syndication partners.
News organizations that ran the material may now face questions about identity checks on freelance contributors. U.S. officials have separately warned that influence actors rotate personas quickly once one cluster is exposed. For readers, the practical takeaway is narrower: conflict coverage that appears suddenly under unfamiliar bylines in multiple small outlets may warrant extra scrutiny when the subject is an active war.