Fake journalist identities placed AI-assisted articles in real news outlets

• OpenAI says the Iran-linked operation’s seven purported journalists placed nearly 100 articles in online publications from July 2025 to October 2026.¹
• A Florida newspaper publisher told The Washington Post he ran one article without checking the author’s identity, trusting the service that sent it.²
• OpenAI described a separate Russia-origin operation in Latin America, involving a purported research centre and deceptive content.¹

An opinion piece about the Iran war reached the Port St. Joe Star, a small Florida weekly, under the name Ervin Hoskins. The paper’s publisher, David Adlerstein, told The Washington Post the article arrived through PeaceVoice, a service that had supplied the paper for years. That existing relationship meant he did not verify the author’s identity, he said.²

OpenAI says Hoskins was one of seven fabricated journalist identities used by an operation whose ChatGPT accounts it assessed as originating in Iran. The company says those users prompted in Persian and used its models to refine long articles, adapt drafts to publications’ submission rules, write pitches and generate social-media comments. That is evidence of AI-assisted production and pitching; it does not establish that every article was written entirely by AI.¹²

The scale is not settled. OpenAI counted almost 100 articles published or syndicated across roughly a dozen outlets, while the Post reported at least 20. The difference may reflect how syndications or publications were counted; the reports do not resolve it.¹²

The articles were not all shown to be factually false. The clearer deception was their claimed authorship: online biographies presented the names as Western journalists, while OpenAI says the bylines belonged to the operation. A June article on CounterPunch, for example, carried a biography describing Hoskins as an American freelance writer and peace advocate. Its publication under that name does not verify who wrote it.¹³

The Florida paper was not the only route into real outlets. The Post reported that Middle East Monitor and Daily Kos removed articles tied to the bylines. NPR, reporting the previous day, said a Hoskins post was still visible in Daily Kos’s community section, where submissions are not reviewed by staff before publication. The accounts describe different outlets and different points in time.²⁴

PeaceVoice’s founder, Tom Hastings, told the Post that the organisation strengthened its existing no-AI policy after OpenAI contacted him. The episode shows how a familiar intermediary can lower the checks applied to an unfamiliar contributor: the Star’s publisher said his confidence in the distributor was why he did not check the writer.²

OpenAI has not identified the Iran operation’s particular actor. It said the activity looked consistent with a commercial, for-hire influence campaign, and that it shared information with relevant authorities. Its account links the banned accounts to Iran, but does not establish that the Iranian government directed the work. The Washington Post disclosed that it has a content partnership with OpenAI.¹²

The Russia-origin case was separate. OpenAI said accounts it linked to Russia used ChatGPT mainly to prepare internal reports and support a purported Latin American research organisation, the Social Research Center. It described efforts to undermine Ukraine and influence politics in Argentina and Bolivia, while saying evidence suggested local employees did not know they were working for a Russian group.¹

Publication is not proof of persuasion. OpenAI said some Iran-linked articles were promoted by outlets’ social accounts, but its assessment concerned potential reach, not a count of readers persuaded or evidence of changed behaviour. The practical breach was earlier in the chain: readers encountered political arguments under identities that, according to OpenAI, were invented.¹

Footnotes

1. OpenAI, “Disrupting AI-enabled ‘false front’ operations” ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

2. The Washington Post, “ChatGPT users in Iran planted AI-generated articles in U.S. news media” ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

3. CounterPunch, “Jared Kushner and the Privatization of America’s Iran Strategy” ↩

4. NPR, “OpenAI caught Russians and Iranians using ChatGPT for influence campaigns” ↩