OpenAI / Wikimedia Commons (Public domain)
The operation, dubbed Bogus Bylines, used seven invented journalist personas and ChatGPT to publish content critical of US military actions in Iran
OpenAI says users in Iran used its chatbot to produce about 100 fake articles that ended up in real publications. The articles carried the bylines of journalists who do not exist.
OpenAI announced the disruption on October 8, 2026. The articles were largely critical of US military actions in Iran, and they landed during a stretch of heightened tension between Washington and Tehran.
How Bogus Bylines worked #
The operators created seven fabricated journalist personas, gave them names and a writing output, and pitched their work to editors.
According to OpenAI, roughly 100 AI-assisted articles were published across a range of outlets between July 2025 and October 2026.
The placements ranged from local US newspapers to international sites. Daily Kos and Middle East Monitor were among the outlets where the content appeared.
Operators wrote their prompts to ChatGPT in Persian, then had the model produce polished English-language copy.
The operators also routed their activity through VPNs. That matters because OpenAI had already restricted direct access to its AI models from Iran, specifically to head off this kind of misuse.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
More than just articles #
OpenAI said the operators also used ChatGPT to draft editorial pitches aimed at getting their pieces accepted.
They also fabricated social media comments. Those comments portrayed the US negatively and cast Iran in a sympathetic light.
OpenAI rated the operation a Category 4 on its six-level influence scale. That is a high score, and the company tied it to how successful the placements were, with articles reaching millions of readers.
OpenAI banned the accounts tied to the operation around October 8, 2026. Some of the outlets that published the content have since removed the articles or the author pages attached to them.
The background: a crowded field #
The same OpenAI report also disclosed the disruption of a similar Russian operation.
OpenAI said activity peaked as US-Iran tensions escalated.
What this means #
Generative models made it easier to produce fluent English copy from Persian prompts at scale. But the articles still had to get past human gatekeepers, and about 100 of them did.
For OpenAI, the disclosure cuts both ways. It shows the company can detect and attribute misuse of its products, down to the language of the prompts and the persona count. It also shows that access restrictions alone did not stop determined users, since the operators reached ChatGPT anyway. The content circulated for more than a year, from July 2025 to October 2026, before the accounts were banned. Some outlets pulled the material only after publication, which means readers had already seen it.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our