OpenAI Says Iran-Origin Accounts Used ChatGPT to Place About 100 Articles Critical of the U.S. War in Roughly a Dozen Outlets
OpenAI banned the accounts and says seven made-up "journalists" carried the bylines. It says it does not know whether the Iranian government was behind them.
Almost 100 articles, seven writers who don't exist
On Thursday, Oct. 8, 2026, OpenAI said it had banned a group of ChatGPT accounts that originated in Iran[1][2]. The accounts supported an operation OpenAI calls "Bogus Bylines." Its core was seven invented "journalists" with Western-sounding names: Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams and Alice Johnson[11].
OpenAI found almost 100 articles under those names. They ran in roughly a dozen small and medium online outlets focused on international affairs, geopolitics and the Middle East[1][11]. The earliest appeared in July 2025 and the latest in October 2026. Output rose after the U.S.-Iran war began this year[12].
Most of the pieces criticized the U.S. war with Iran. OpenAI summarized their framing as casting the U.S. as the aggressor and Iran as a defiant victim[2].
What the AI actually did
The operators mostly prompted the model in Persian, also called Farsi. They then used it to edit English-language articles, draft pitch emails to editors and write social media comments[2][6]. They also tailored pieces to each outlet's submission rules[2].
OpenAI's argument is that the danger here is not AI writing propaganda from scratch. It is that AI makes the dull work faster: translating, polishing and fitting a pitch to each outlet[2]. A plausible freelancer in fluent English now costs very little to produce.
The outlets named include the progressive U.S. site Daily Kos, the U.K.-based Middle East Monitor and a small Florida paper, the Port St. Joe Star[4][3]. The Star got its Hoskins piece through PeaceVoice, a syndicator that had sent it commentary for years[12]. Contributor sites and syndicators rarely verify who their writers are. That trust is how the system works, and it's the gap this operation used[12][4].
Who ran it? Nobody knows
Here is the tension at the center. Some headlines said "Iranian operatives" planted "fake stories" in "US media outlets"[2]. OpenAI's own report says something narrower. It describes the accounts as Iran-origin. It says it is unclear who ran the operation or whether the operators are tied to Iran's government[2].
CNN's headline used the firmer wording, though its body carried the caveat. It also noted that none of the outlets appears to be a major mainstream publisher[2]. Some do reach large audiences, though. Middle East Monitor has almost two million Facebook followers[2].
The word "fake" also needs care. The bylines were invented. OpenAI did not say the articles contained invented facts. They were opinion pieces under made-up names.
Why OpenAI is telling this story
Nearly every fact here comes from OpenAI's own report, so its choices set the frame. The company argues that AI firms can see what no one else can. Only the model provider sees the prompts, so only it can spot the pattern, cut the accounts off and warn outlets[1][5].
It also has reasons to be seen doing this. Lawmakers are weighing AI rules, and the company is seeking national-security business. Regular threat reports also discourage abuse[1][7]. The disclosure carries a cost, too. It concedes that OpenAI's tool helped an influence operation reach real publications[2].
OpenAI rated the operation Category 4 on a six-point "breakout" scale. The scale measures how far a campaign spreads beyond the fake accounts that started it. A low score means the content stayed in its own bubble. Category 4 means it got into real publications[6][11]. It does not measure how many readers saw the pieces or changed their minds. No one has measured that.
In the same report, OpenAI banned a Russia-origin operation it calls "Dark Clark." That group used a fake think-tank leader to push stories in Latin America[4][5].
Four ways to read it
Security-minded critics, including Republican-leaning outlets and officials, see a wartime adversary shaping American debate. In their view, someone posed as Americans to argue against a war the U.S. is fighting. They say hiding who is speaking is the harm. Readers judge an argument partly by who makes it, and a fake "American freelance writer" removes that check[7][3]. They want tougher platform policing and closer vetting at outlets[2][7].
The outlets say they were deceived, not complicit. Daily Kos took down the fake authors' pages[4]. According to NPR, Middle East Monitor removed at least 21 articles by four of the fake authors[3]. Small papers face the same risk with fewer resources to check identities[12].
Tehran's side is hard to state, because no response from Iran's government turned up in the coverage reviewed. OpenAI made no state attribution[2]. Anyone defending Iran would likely start there: a private company has not shown who the operators were, and a state shouldn't be blamed without proof. If the operators were state-linked, a motive would probably be to shape opinion at home and abroad. OpenAI says it can't determine that.
Iran International, a London-based Persian-language channel opposed to Iran's government, stresses a different part of the report. OpenAI found more than two dozen batches of hostile Persian replies aimed at its posts. It reads this as part of a pattern of pressure on Persian-language media outside Iran[6].
Same report, different headlines
Coverage split mostly on how much of OpenAI's caution survived. The Hill used plain location wording, "users in Russia, Iran," and avoided asserting a state link[4]. NPR said OpenAI "caught" Russians and Iranians, which casts the company as the hero and swaps "origin" for nationality[3]. The Washington Examiner adopted OpenAI's own security language, "false front" and "threat actors," and gave the attribution caveat little weight[7].
Outside the U.S., India's WION called the content "propaganda" in its own voice[9]. The Sunday Guardian hedged with "suspected" but ran the story inside a war live blog, tying it to the battlefield[8]. Iran International led with itself as the target, and it misdated the report to Oct. 9[6].
One number also varies. This story's original assignment said "at least 20 outlets." OpenAI's report and CNN both say roughly a dozen[1][2].
What's left open is large. The record shows the deception. It does not show who was behind it, how many people read the pieces, or whether any changed their views. Some will see foreign deception of American readers in wartime. Others will see fake bylines on opinions many Americans already hold. Banning accounts on one platform also doesn't stop the same people from using other tools.
Summary
OpenAI said on Thursday, Oct. 8, 2026, that it had banned a group of ChatGPT accounts that originated in Iran[1][2]. It says the users invented seven Western-sounding "journalists." They used ChatGPT to polish articles, fit them to each outlet's submission rules, and write pitch emails to editors[2]. OpenAI found almost 100 articles published or syndicated under those bylines in roughly a dozen online outlets[1][11]. Most of the articles criticized the U.S. war with Iran[2]. Named outlets include the progressive U.S. site Daily Kos, the U.K.-based Middle East Monitor, and a small Florida paper, the Port St. Joe Star[4][3].
None of the outlets appears to be a major mainstream publisher, CNN reported. Some do have large audiences. Middle East Monitor has almost two million Facebook followers[2]. Outlets that were named have removed the fake authors' pages or articles[4][12]. In the same report, OpenAI also banned a Russia-origin operation it calls "Dark Clark." That group used a fake think-tank leader to push stories in Latin America[4][5].
The biggest open question is who ran the Iranian operation. OpenAI says it does not know whether the operators are tied to Iran's government[2]. Some headlines still called them "Iranian operatives"[2]. A second question is what kind of harm was done. The articles were opinion pieces under fake names. OpenAI did not say they contained invented facts. Their framing cast the U.S. as the aggressor and Iran as a resilient victim, according to OpenAI[2]. So reasonable people can disagree about the problem. Some will see foreign deception of American readers in wartime. Others will see a fake byline on views that many Americans also hold. The record shows the deception. It does not settle how much the articles changed anyone's mind.
The Event
On Oct. 8, 2026, OpenAI published a threat report saying it had banned ChatGPT accounts originating in Iran that ran an operation it named "Bogus Bylines"[1][2]. The accounts mostly prompted the model in Persian (Farsi). They used it to edit English-language articles, draft pitches to editors and write social media comments[2][6]. OpenAI says almost 100 articles ran under seven fake bylines in roughly a dozen outlets between July 2025 and October 2026[1][11][12].
Undisputed Facts
- OpenAI announced on Oct. 8, 2026, that it had banned accounts tied to an Iran-origin operation it calls "Bogus Bylines" and a Russia-origin operation it calls "Dark Clark"[1][4][11].
- The Iranian operation used seven invented personas: Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams and Alice Johnson[11].
- OpenAI identified almost 100 articles under those bylines in roughly a dozen small to medium online outlets focused on international affairs, geopolitics and the Middle East[1][11].
- Outlets that ran the articles include Daily Kos, Middle East Monitor and the Port St. Joe Star[3][4].
- The earliest article OpenAI identified ran in July 2025. The latest ran in October 2026. Output rose after the U.S.-Iran war began this year[12].
- OpenAI said it is unclear who ran the operation or whether it is affiliated with the Iranian government[2].
- OpenAI said the operation also produced more than two dozen batches of hostile Persian-language replies to posts by Iran International[6].
The Pressure
Strip away the moralizing and blame. What structural realities persist regardless of which narrative wins?
- Wartime battle over U.S. opinion
- With the U.S. at war with Iran, American public support is a strategic asset. Output from this operation rose after fighting began[12]. That fits an effort to shift U.S. and global opinion, whoever ran it[2].
- Open publishing runs on trust
- Contributor sites and op-ed syndicators rarely verify identities. A polished pitch from a plausible "freelancer" can pass through PeaceVoice to a Florida weekly, or onto Daily Kos[12][4]. AI cuts the cost of producing polished pitches in fluent English[2].
- AI firms as de facto watchdogs
- Only the model provider can see the prompts. So public knowledge of this operation depends on OpenAI choosing to disclose it, and on its own judgment about attribution[1][5].
Material realityAlmost 100 articles under seven invented bylines ran in roughly a dozen small and medium outlets over about 15 months[1][11][12]. The pages have been removed where outlets acted[4][12]. OpenAI rated the effort Category 4 of 6, meaning it got into real publications[6]. But no one has measured how many people read the pieces or whether they changed anyone's views. The operators have not been identified, and OpenAI did not link them to the Iranian state[2]. Banning accounts on one AI platform does not stop the same people from using other tools.
Narrative as a weaponOpenAI is the main shaper of this story. Nearly every fact comes from its own report, so its choices about wording, scale ratings and attribution set the frame. It wants to look like a responsible guardian of its tools. U.S. outlets on both sides largely adopted that frame. Some headlines then went past OpenAI's own caveats, using phrases like "Iranian operatives" and "fake stories" in "US media." Iran International stressed the harassment of diaspora media. Iran's government stayed silent in the coverage reviewed. One number needs care: this story's original assignment said "at least 20 outlets," but OpenAI's report and CNN both say "roughly a dozen"[1][2].
How Each Side Sees It
Each major actor’s view — how it frames things, its underlying incentive, and how it’s materially affected. Tap a side to read it.
Frames it asOpenAI argues that AI companies can see things no one else can. They can watch how bad actors use their tools, cut them off, and publish what they find so outlets and the public can defend themselves[1]. Its report says the danger is not that AI writes the propaganda. The danger is that AI makes the dull work faster: translating, polishing and fitting pitches to each outlet's rules[2]. OpenAI also argues that article placement is worth noticing because it reached more people than fake social media accounts usually do[12]. It rates this operation Category 4 on a six-point "breakout" scale[6][11]. That scale measures how far a campaign spreads beyond the fake accounts that started it. A low score means the content stayed inside its own bubble. A higher score means real outlets or real people picked it up and passed it on. Category 4 means the content got into real publications. It does not, by itself, measure how many readers changed their minds.
WhyOpenAI wants to show governments and the public that it polices its own product. That matters as lawmakers weigh AI rules and as the company seeks national-security business. Regular threat reports also discourage abuse[1][7].
Impact on themDisclosure is good publicity. It also concedes that its tool helped an influence operation get printed in real outlets[2].
Frames it asThese outlets were deceived, not complicit. Many open-contribution sites and small papers depend on outside writers and syndication services. Those systems run on trust. The Port St. Joe Star got its Hoskins piece through PeaceVoice, a syndicator that had sent the paper commentary for years[12]. Once told, outlets acted. Daily Kos took down the fake authors' pages[4]. Middle East Monitor removed articles under the flagged names[12].
WhyThey want to protect their credibility and keep open submission systems working without becoming a channel for state-linked deception[4][12].
Impact on themAccording to NPR, Middle East Monitor removed at least 21 articles by four of the fake authors[3]. The episode has brought scrutiny of how open-contribution sites check contributors[4]. Small papers face the same problem with fewer resources to check identities[12].
Frames it asIn their view, the core issue is a wartime adversary secretly shaping American debate. A foreign actor posed as Americans to argue against a war the U.S. is fighting. That is deception whatever one thinks of the war[7][2]. They argue that hiding who is speaking is the harm. Readers judge an argument partly by who makes it. A fake "American freelance writer" takes that judgment away[3].
WhyThey want to keep public support for U.S. policy toward Iran and to push platforms and outlets to treat foreign operations as a security threat[7].
Impact on themThe finding strengthens calls for stronger platform policing and tighter vetting at outlets during the war[2][7].
Frames it asNo response from Tehran turned up in the searched coverage. OpenAI's report does not tie the operation to the state[2]. OpenAI summarized the content as portraying the U.S. as the aggressor and Iran as a defiant victim[2]. Anyone defending Iran would likely start from a simple point. A private company has not shown who the operators were, and the Iranian state should not be blamed without proof.
WhyIf the operators were state-linked, which OpenAI says it cannot determine, the aim would likely be to weaken U.S. domestic support for the war and shape global opinion. Absent attribution, Tehran's stated interest is to reject the accusation.
Impact on themThe finding adds to the record of Iran-origin influence work, though OpenAI made no state attribution. It may make outlets more suspicious of pieces that echo Tehran's line[1][2].
Frames it asIran International says the report confirms that it is a target. OpenAI found that the same operation repeatedly produced hostile Persian replies to its posts[6]. It treats the case as part of a pattern of pressure on Persian-language media outside Iran[6].
WhyTo document hostile campaigns against itself and its audience[6].
Impact on themIts reporting puts the harassment of diaspora media at the center, an angle most U.S. coverage left out[6][2].
Like this article?
The Bias Ledger average rating 3.7
The same story, as framed by outlets across the spectrum, ordered least to most biased. The bias score (1 = straight, 10 = heavily spun) is an AI assessment of that framing — click an outlet to see its track record. The tell is the word choice or omission that reveals the angle.
| Outlet | Vantage | Bias | How they frame it | The tell |
|---|---|---|---|---|
| The Hill | U.S. center | 2 | OpenAI users in Russia, Iran used models to push political agenda online, in media | It uses plain, location-based wording ("users in") and avoids asserting a state link. It names Daily Kos and the takedown of author pages. |
| NPR | U.S. center-left | 3 | OpenAI caught Russians and Iranians using ChatGPT for influence campaigns | "Caught" casts OpenAI as the hero. Nationality labels stand in for "Russia-origin" and "Iran-origin" accounts. The body gives concrete detail, such as the Hoskins byline in specific outlets. |
| Washington Examiner | U.S. right | 3 | OpenAI bans 'false front' influence operations in Russia and Iran | The headline puts OpenAI's own term in quotes. The lead adopts the company's security language ("cracked down on threat actors"). The attribution caveat gets little weight. |
| Iran International | London-based Persian-language broadcaster opposed to the Islamic Republic | 4 | Iran-origin campaign targeted Iran International, foreign media - OpenAI | It makes itself the lead target ahead of the outlets that actually ran the articles. It accurately keeps OpenAI's "Iran-origin" wording but misdates the report to "Thursday, October 9." |
| The Sunday Guardian | Indian (right-leaning weekly) | 4 | OpenAI Uncovers Suspected Iran-Linked Campaign That Planted Fake Articles In US Media | "Suspected" is a careful hedge. But "fake articles" and "US media" overstate what was shown. It ran inside a war live blog, which ties the story to the battlefield. |
| CNN | U.S. center-left | 5 | Iranian operatives used AI to plant fake stories in multiple US media outlets | "Iranian operatives" suggests a known actor, but the body says OpenAI cannot say who ran it or whether Iran's government was involved. "Fake stories" blurs fake bylines with fake facts. "US media outlets" narrows a global set of outlets. The body does carry the caveats. |
| WION | Indian (private broadcaster, nationalist-leaning) | 5 | OpenAI shuts down Russian, Iranian influence operations using ChatGPT to spread propoganda | "Propaganda" appears in the outlet's own voice, not attributed. "Russian, Iranian" suggests national ownership that OpenAI did not claim for Iran. |
References
- Disrupting AI-enabled "false front" operations — OpenAI · Primary source; the company that operates ChatGPT and has a commercial and policy interest in showing it polices misuse
- Iranian operatives used AI to plant fake stories in multiple US media outlets — CNN · U.S. commercial cable news; center-left
- OpenAI caught Russians and Iranians using ChatGPT for influence campaigns — NPR (via KPBS) · U.S. public radio, funded by member stations, donors and underwriting; center-left
- OpenAI users in Russia, Iran used models to push political agenda online, in media — The Hill · U.S. political trade outlet owned by Nexstar; center
- OpenAI says Iran, Russia used AI journalists, think tanks to influence Western media — CyberScoop · U.S. cybersecurity trade publication (Scoop News Group); industry-focused
- Iran-origin campaign targeted Iran International, foreign media - OpenAI — Iran International · London-based Persian-language broadcaster opposed to the Islamic Republic; funding has been reported to trace to Saudi-linked backers, which it disputes
- OpenAI bans 'false front' influence operations in Russia and Iran — Washington Examiner · U.S. conservative outlet owned by Clarity Media (Anschutz)
- US-Israel-Iran War Latest Live News: OpenAI Uncovers Suspected Iran-Linked Campaign That Planted Fake Articles In US Media — The Sunday Guardian · Indian weekly; right-leaning, generally supportive of the BJP government
- OpenAI shuts down Russian, Iranian influence operations using ChatGPT to spread propoganda — WION · Indian private broadcaster (Essel Group); nationalist-leaning
- OpenAI says Iran, Russia influence ops busted — Al-Monitor · U.S.-based Middle East news site, privately owned; carries wire copy
- OpenAI Bans Two Covert Influence Operations Using False Fronts — Unite.AI · AI-industry news site; generally pro-technology
- US-Israel-Iran War Latest Live News: Iran-Linked Network Used 'Fake Journalist' Personas to Place US-Iran War Articles in Online Publications Around World, Says OpenAI — The Sunday Guardian · Indian weekly; right-leaning, generally supportive of the BJP government