OpenAI Busts Fake Journalist Rings Pushing AI Propaganda

OpenAI says it has banned two covert influence operations that used ChatGPT to power rings of Fake Journalist personas — fabricated reporters whose planted stories ended up in real news outlets around the world. The bans, disclosed this week, target a Russia-linked network and an Iran-linked effort that laundered geopolitical messaging into legitimate media. One of the phony personas fronted a fake Latin American think tank, and together the two operations placed almost 100 articles in small-to-medium publications, according to OpenAI's report. It is a reminder that the propaganda landing in your feed can look exactly like ordinary news.

The Russian-linked effort was the more severe of the two. OpenAI rates it at the highest tier of its six-point severity scale — the first operation to hit that level in two and a half years of public disclosures — because the planted content drew responses from elected officials. The Iran-linked effort, meanwhile, built seven fake reporter profiles posing as journalists from Western countries, including the United States, and placed long-form articles across roughly a dozen online outlets, according to AFP's reporting.

How the Fake Journalist Rings Worked

The Russian operation, which OpenAI calls Dark Clark, centered on a false-front research platform named the Social Research Center in Latin America. OpenAI says the group controlled the outfit through an invented persona named Mia Clark and roped in local staffers who, by all available evidence, had no idea who was really running the show. The accounts used ChatGPT mostly to draft internal status reports for unseen superiors — and those same reports helped investigators map the network. The planted material ranged from fabricated audio files and documents to staged political stunts that triggered fact-checks and official denials in Ecuador and Peru.

The Iran-linked operation, which OpenAI calls Bogus Bylines, worked differently: its Fake Journalist bylines — each presented as a Western reporter — were used to pitch and polish long-form articles for placement in outlets covering international affairs, geopolitics, and the Middle East. One persona, an alleged American freelance writer named Ervin B. Hoskins, ran social accounts on major platforms that were later suspended, as reported by NPR. The earliest planted piece dates to July 2025, the pace picked up through 2026, and the operation was still active this month.

Notably, ChatGPT was not mainly churning out the propaganda itself. OpenAI found the Russian accounts leaned on the chatbot for internal reporting, while the Iranian operators used it to polish English drafts and adapt messaging across languages, as reported by The Decoder. That made the fake copy read more fluently — and harder for editors to spot. Detecting Fake Journalist content, in other words, is getting harder precisely because the language is getting better.

Why the Fake Journalist Busts Should Matter to You

For young readers, the stakes are simple: the fake stories did not live on sketchy fringe sites. They ran in real publications — one outlet that carried the planted pieces had nearly two million Facebook followers, according to the investigation. If you scroll news on social apps, a Fake Journalist byline could already have passed through your feed without a single red flag.

A ban is a start, not a solution. OpenAI itself found that campaigns landing stories in established media reached wider audiences than fake-social-account campaigns, according to NPR's reporting — which is exactly why this tactic keeps coming back. Researcher Darren Linvill, who studies online influence operations, told NPR the AI helped operatives produce language that "looked and felt and read authentic" enough for legitimate outlets to publish it.

The Fake Journalist Playbook, Old and New

This is not the first time phantom reporters have fooled real newsrooms. OpenAI compared the new cases to pre-AI fronts like "Alice Donovan," a fake journalist whose articles ran in Western outlets in 2016 and 2017, and "PeaceData," a bogus outlet that recruited unwitting writers in 2020, according to Unite.AI. The AI-era difference is scale and fluency: the company says it has now exposed 30 covert influence operations since it started publishing these reports.

The takeaway for media-literate readers is not paranoia — it is verification. Check the byline, check the outlet's corrections page, and treat unsourced viral claims with suspicion even when the branding looks credible. The Fake Journalist campaigns were caught because the operatives' own ChatGPT-written reports left a trail, and because investigators followed it. Your feed deserves the same skepticism you would give a too-perfect stranger online.

Related reading: OpenAI Fired Researchers Say Safety Was the Reason — and more on our AI News topic page.