OpenAI Bans Russia and Iran False-Front Influence Operations
OpenAI on 8 October 2026 banned Russia- and Iran-origin ChatGPT account clusters behind false-front influence campaigns it rated Category 5 and Category 4 on its breakout scale.
PromptCrates Editorial
Staff Writer
OpenAI said on Thursday, 8 October 2026, that it banned clusters of ChatGPT accounts traced to Russia and Iran for supporting sophisticated "false front" influence operations that mixed AI drafting with older tradecraft. OpenAI's report describes a Russia-origin campaign nicknamed Dark Clark that co-opted unwitting Latin American staff into a "think tank," and an Iran-origin campaign nicknamed Bogus Bylines that pitched long-form articles under seven fake journalist personas. NPR reported the same day that OpenAI described one Russian campaign as the widest-reaching influence operation it has come across so far, after it drew a public response from a member of the European Parliament.
What OpenAI says Dark Clark did
According to OpenAI, Russia-origin operators used VPNs to reach ChatGPT and prompted mainly in Russian, with one Spanish-speaking operator still appearing to be located in Russia. Their heaviest use of the model was drafting internal reports to a superior about campaigns across Latin America, including efforts to undermine Ukraine's reputation and to interfere in local politics, especially in Argentina and Bolivia. OpenAI says the operators also controlled a self-styled research platform called the Social Research Center through a fake persona named Mia Clark, and that on-the-ground employees appeared unaware they were working for a Russian group.
OpenAI assessed Dark Clark as Category 5 on the IO Breakout Scale, citing evidence that some fakes prompted fact checks, official denials, and public comment by politicians. One example the company says it corroborated involves a May 2026 fake email purporting to come from Peru's Regional Directorate of Education in Lima, instructing schools to hold Ukraine-related events that referenced Stepan Bandera; matching stories later appeared in Peruvian and Polish press and drew comment from a Polish Member of the European Parliament, OpenAI wrote. Other internal claims described fake audio and video campaigns in Ecuador and Bolivia that open-source researchers or officials later debunked, according to the report.
OpenAI is careful about impact inflation. It notes that operators sometimes took credit for events they did not cause, and that some claimed fakes never appeared in open sources. The company still calls Dark Clark the most complex front-identity attempt it has disrupted in two and a half years, partly because the Social Research Center produced a majority of original research content via co-opted staff rather than simple plagiarism. OpenAI says it shared information with relevant authorities.
How Bogus Bylines planted articles
The Iran-origin cluster prompted in Persian and generated content in Persian and English, also via VPNs, OpenAI said. Operators used ChatGPT to refine English-language articles about the US-Iran conflict, check drafts against outlet submission criteria, and draft pitch emails. Seven bylines appeared in OpenAI's investigation: Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams and Alice Johnson. Personas were backstopped on social platforms; transparency settings for some accounts pointed back to Iran, the report says.
OpenAI identified almost 100 articles published or syndicated under those bylines across roughly a dozen small-to-medium outlets between July 2025 and October 2026, with publication density rising after the US-Iran conflict intensified. That article-planting track is rated Category 4. A parallel mass-commenting track, generating batches of English and Persian replies, is rated only Category 2 because engagement was low and many likes came from accounts linked to the operation itself. Internal "impact" metrics that counted views on parent posts rather than on the replies exaggerated performance, OpenAI wrote. Activity appeared consistent with a commercial for-hire actor, though OpenAI says it could not identify the specific contractor.
Why false fronts matter for AI policy
OpenAI places both campaigns in a longer pattern: of about 30 covert influence operations it has exposed since early 2024, those that land content in real media tend to score higher on reach than those that rely only on fake social accounts. The company compares Bogus Bylines to earlier fake-journalist fronts such as Alice Donovan, and Dark Clark to PeaceData's use of unwitting writers. AI, in OpenAI's framing, mainly cheapens drafting, translation, pitching and internal reporting rather than inventing a wholly new influence model.
That policy angle connects to earlier PromptCrates coverage of threat disclosure and platform enforcement, including Anthropic's September threat intelligence report, the FTC probe of OpenAI and Anthropic, and Australia's Senate calls for testimony on rogue agents. Thursday's OpenAI post is an enforcement-and-transparency story: account bans plus a detailed public write-up intended to make further research and disruption easier.
NPR's account emphasises the same widest-reach finding for the Russian campaign and notes that Iranian operators used AI to polish articles and pitches while Russian operators leaned on ChatGPT for status reports to bosses. Both outlets stress that not all published content was model-generated; AI sat inside a wider workflow of fake personas, cutouts and media pitching.
What remains unsettled after the bans
OpenAI does not claim the campaigns are unique to ChatGPT, and it does not publish a full list of every outlet that ran Bogus Bylines copy. Attribution for the commercial Iranian actor remains unresolved in the public report. Dark Clark's ties to publicly discussed "Politology" or "La Compania" activity are presented as overlaps from open-source research and operator self-reporting, not as a courtroom finding. Readers should also remember OpenAI's own caveat that operator boasts are unreliable without corroboration.
Still, the dated facts for 8 October 2026 are concrete: OpenAI banned the Russia- and Iran-origin account clusters, published a long disruption report, rated Dark Clark Category 5 and Bogus Bylines Category 4 on the article-planting track, and said nearly 100 articles appeared under the Iranian personas. Newsrooms, platforms and regulators tracking AI-assisted influence now have a primary-source case study of false fronts that look like pre-AI campaigns with faster drafting. Follow-on questions for policy teams include how publishers verify freelance pitches, how VPN bans interact with genuine diaspora users, and whether other model providers will match OpenAI's level of public detail when they disrupt similar clusters.
