OpenAI disrupts two AI-enabled false front influence operations, with the Russia-origin campaign reaching Breakout Scale Category 5 for the first time

Start with the headline and the assessment. Per OpenAI's official post: it banned two operations, one Russia-origin and one Iran-origin, that used ChatGPT in combination with more traditional techniques to support sophisticated false front entities, then used those entities to launder geopolitical and conflict-related messaging into target audiences. The Iranian operation included a stable of seven journalist personas used to pitch long-form articles to small and medium online outlets around the world; the Russian operation appears to have co-opted unwitting people in Latin America to run a think tank on the ground. In OpenAI's assessment the Russia-origin operation belongs in Breakout Scale Category 5 — the first Category 5 operation the company has disrupted since it began reporting — while the Iran-origin operation reached Category 4. OpenAI also notes something worth pausing on: what is most striking about these operations is that they closely resembled complex influence operations of the pre-AI age, but used AI to make some of the workflows easier.

The Russia case: a think tank under someone else's control. OpenAI banned a cluster of ChatGPT accounts originating in Russia that used the models for a range of covert influence tasks targeting countries across Latin America; much of the activity appeared aimed at undermining Ukraine's reputation in the region, with some aimed at influencing local political outcomes, especially in Argentina and Bolivia. Most operators prompted in Russian, one prompted in Spanish but nevertheless appeared to be located in Russia, and because OpenAI does not allow access from Russia they used VPNs. They used ChatGPT for three main tasks: writing internal reports to an unknown superior, creating content for their operations, and drafting reports on the performance of a self-described research platform in Latin America called the Social Research Center (SRC). They controlled the SRC through a fake persona named Mia Clark, which is why OpenAI nicknamed the operation Dark Clark. The reasoning in the report is blunt: the operators' reports referenced decisions over pay scales, hiring and firing, suggesting they controlled the entity rather than cooperating with it, and the available evidence indicates the SRC's employees in Latin America were not aware they were working for a Russian group. OpenAI identified well over 60 articles on the SRC website, the great majority appearing to be original compositions — which sets Dark Clark apart from another Russia-linked think tank front OpenAI recently exposed that plagiarized most of its content, and suggests an intention to build a more durable front entity.

What the Russia case actually faked. Per the official report and the open-source work in it: in May 2026 the operators said they created a fake email address purporting to come from the regional education directorate in Lima, Peru, and used it to instruct schools in the district to hold events dedicated to Ukraine on the national Day of Cultural and Linguistic Diversity on May 21, with explicit instructions to reference controversial figures including Stepan Bandera. According to the operators, some schools replied to the fake address, confirming they had held such events and even providing pictures. They then claimed to have planted stories in the Peruvian and Polish press alongside allegations that Ukraine was exporting ultra-nationalist ideologies and triggering outrage; open-source searches identified matching coverage in the Peruvian and Polish press and an English-language publication in Hungary, some of which have since been deleted, including one carrying a Polish MEP's proposal that people showing anti-Polishness be declared persona non grata. In June the operators said they used a different fake address to trick schools in Ecuador into holding a ceremony pledging allegiance to President Daniel Noboa and Erik Prince, former head of Blackwater, reportedly provoking outrage and pressure on the government to deny it; open-source research found matching press coverage and a detailed rebuttal by Ecuador's Minister for Education. In March they spread two fakes targeting Ecuador: audio attributed to Ukraine's consul there making disparaging comments about Ecuadorians, and a fake video clip carrying a real news channel's logo alleging Noboa's government was recruiting young men to fight in Ukraine. In late May, at the height of anti-government protests in Bolivia, they spread fake audio of a worker at the state water company EPSAS saying the government would shut off water to La Paz and declare a state of emergency — the report states the campaign's explicit goal was to exacerbate the crisis around the protests. Earlier still, in 2024 they said they ran two campaigns targeting Argentina's President Javier Milei: spreading a false report that he had bought Cartier jeweled collars for his dogs, and paying local actors to post anti-Milei graffiti in Buenos Aires. Based on leaked documents, both have been publicly attributed to entities known as Politology and La Compania.

The Iran case: seven bylines, nearly a hundred articles. OpenAI banned a cluster of ChatGPT accounts originating in Iran that prompted in Persian and generated content in Persian and English, obscuring their location with VPNs. The operation used the models to refine long-form articles and editorial pitches, generate social media comments, and polish internal reports. It pitched under seven different bylines, hence the nickname Bogus Bylines: Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams and Alice Johnson. According to the biographies posted alongside some of the articles, each persona purported to be a Western journalist, usually American though the Noah Lamington bio claimed New Zealand, often focused on international politics, the Middle East, human rights or conflict. For the Michael Harrison persona the operators asked the model to write the bio itself. Using open-source investigation OpenAI identified almost 100 articles published or syndicated under those bylines across more than a dozen outlets, generally focused on international affairs, geopolitics and events in the Middle East, sometimes from an antiwar or progressive point of view; the earliest identified article was published in July 2025 and the latest in October 2026, with the frequency rising sharply after the outbreak of the US-Iran conflict. On comments, the typical approach was to feed the model a social media post or a screenshot and ask for a set of replies aligned with the operation's focus, then post them within minutes of each other — the screenshots in the report show one batch of English comments posted in a 10-minute span and a batch of Persian comments posted within 13 minutes. The operators generated over two dozen batches of hostile Persian-language replies to posts by the UK-based satellite network Iran International, a critic of the Iranian regime.

How big the impact was, in OpenAI's own split accounting. The Iranian operation's two workstreams differed sharply: the social media replies OpenAI could identify generally enjoyed low numbers of likes, views and comments, in the single or double digits, and typically made up a minority of total comments on the post in question, which OpenAI assesses at Breakout Scale Category 2; the journalist personas were a different story, with almost 100 articles landing in outlets that include some with substantial followings — the report notes one such outlet had almost 2 million followers on Facebook, almost 355,000 on X and over 544,000 on Instagram as of August 2026, and assesses this part at Category 4. One detail shows how unreliable the operators' own measurement was: their main impact metric was the number of views on the posts they replied to rather than on the replies themselves, counted before and after and without accounting for views that came from the original posts' own popularity; OpenAI states plainly that this greatly exaggerated the operation's impact figures. The Russia case carries the same caveat: OpenAI warns against taking the operators' internal claims of impact at face value, because one of their main approaches was to take credit for activity that had nothing to do with them — including claiming that Argentina's foreign minister quoted their operation in a UN committee, and claiming a Brazilian media report about a captured Brazilian reproduced their messaging when in fact it quoted the captive's own video. OpenAI describes that as an attempt to run an influence operation targeting their own employers.