Iran-based campaign used ChatGPT to pitch editors under seven fake bylines

The publisher of the Port St. Joe Star took one of the campaign's articles from PeaceVoice, an organization his paper had worked with for years, according to the Washington Post as summarized by Digital Trends ^[14]. He said he did not independently verify the writer's identity or the article's origin because of that relationship ^[14]. Some outlets have since removed articles, and PeaceVoice has tightened its AI policy, the Post reported ^[15].

Here's what contributions desks tend to tell themselves: the AI risk sits in the copy, so a detector covers it. Here's what the operators actually did. They asked ChatGPT to review English drafts against a given publication's submission requirements and suggest edits, then had it write the email pitching the editor ^[12]. ChatGPT wrote at least one persona's biography, and some personas had social accounts to back them up ^[13]. Each step was about making the submission look like it belonged at the outlet receiving it ^[12]. "AI detection won't solve this by itself," Digital Trends wrote ^[16].

Volume helped as well. Almost 100 articles across seven personas comes to about 14 per invented writer ^[19], spread over the 15 months between the earliest piece in July 2025 and the latest in October 2026 ^[20]. I'd expect a name with a dozen-plus clips on one beat to pass the quick search most editors run before saying yes. The people who have to act on this run contributor and syndication desks at outlets of the kind OpenAI listed. "These were small to medium outlets, generally focused on international affairs, geopolitics, and events in the Middle East," OpenAI said ^[4].

The cheapest check in the record came from the platforms. One persona, Ervin B. Hoskins, had a bio describing him as "an American freelance writer" and accounts on X and Instagram ^[7]. X's transparency information showed the account connecting through a "West Asia android app", and Instagram's showed it based in Iran, according to screenshots OpenAI provided ^[8]. The labels contradicted the bio. Both accounts appear to have been suspended ^[8].

The Russian operation is harder to catch at the byline. "The SRC appears to have produced a majority of original content via its co-opted staff. This stands out as the most complex attempt to run a front identity that we've disrupted over the past two and a half years," OpenAI said of the Social Research Center ^[10]. A writer check there confirms a real person working at a real-looking think-tank. For that case, the useful question is who runs the organization and who pays for it.

Sort a contributor queue on two lines. One is how the piece arrived: cold, or through someone the desk already trusts, such as a syndicator or a long-time partner. The other is where the writer's identity comes from: something the writer did not send, such as a phone call or a platform's account-location label, or only the bio and clips in the pitch. A cold pitch with a self-supplied identity at least looks like a stranger. The quadrant that needs staff time is a trusted channel carrying a self-supplied identity, and the Port St. Joe Star article sat in it ^[14].

I'd have syndicators verify identity once, upstream, and tell the outlets they supply that they did, so neither end assumes the other has checked. The tradeoff is editor time, plus some genuine freelancers with thin online records who get slowed down or dropped. That cost falls on the same small and medium outlets OpenAI said the campaign reached ^[4].