A Campaign From Iran, With No Named Sponsor
OpenAI called the campaign Bogus Bylines. Operators prompted ChatGPT in Persian, generated Persian and English material and used VPNs to conceal their location. The company banned the account cluster and shared information with relevant authorities.
Its assessment stopped short of identifying an Iranian government agency. OpenAI said the activity appeared consistent with a commercial actor conducting an influence campaign for a client, but it could not identify the particular actor. The report did not establish who commissioned or funded the operation.
The Israel angle appears in the requested social comments: users wanted messages portraying Israel as an unreliable US ally, the United States as the aggressor and Iran as a defiant, resilient victim. These were campaign narratives, rather than OpenAI's findings about the conflict.
Drafts Tailored to Editors
The operators asked ChatGPT to compare English article drafts with specific publishers' submission requirements and suggest changes. After approving revisions, they requested an email to pitch the article to the relevant editor.
That workflow went beyond generating political posts. It used existing editorial channels and the credibility of apparently independent contributors to reach readers. OpenAI did not say that its models wrote every article or that every assertion in the published material was false.
The seven bylines were Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams and Alice Johnson. Their biographies presented them as Western journalists, usually American. For Harrison, OpenAI observed a request to generate the biography itself.
Fake Profiles, Real Publication Channels
Some identities had supporting social accounts. Hoskins claimed to be an American freelance writer, while platform transparency information pointed to Iran. Such inconsistencies formed part of OpenAI's evidence linking the personas to the operation.
OpenAI described the publishers as small to medium outlets covering international affairs, geopolitics and the Middle East. The earliest article it identified appeared in July 2025 and the latest in October 2026. Publication frequency increased after the conflict began.
Its report used both roughly a dozen outlets and more than a dozen publications. The almost 100 total includes articles published or syndicated; it is not a count of 100 original submissions. Some publishers promoted the articles through their own social accounts.
Manufactured Praise, Limited Response
Alongside editorial pitches, the operators generated batches of English and Persian replies. Some praised their own articles after publishers shared them. OpenAI also found requests for more than two dozen batches of hostile Persian replies to Iran International.
The company located only a subset of the generated comments. Those found generally attracted likes, views and comments in single or double digits, and some likes came from accounts linked to the operation itself.
OpenAI rated the commenting activity Category 2 on the six category Breakout Scale, while article placement reached Category 4. The latter reflects repeated distribution through media outlets, without establishing readership or persuasion.
What the Disruption Leaves Unanswered
The operators' internal reports counted views of the original posts they replied to as evidence of their own impact. OpenAI said that method exaggerated effectiveness because it did not isolate views of the replies or other causes of increased traffic.
For readers assessing influence operations, the distinction matters: deceptive authorship and successful placement are established findings; a measurable change in public opinion is not. The original report does not provide a complete outlet list or establish how many articles were removed. The operator, client and funding remain unidentified.
