Disrupting a new covert influence campaign from Russia is a threat-intelligence report published by OpenAI's Global Affairs team on August 25, 2026 (Source: openai.com). It describes the banning of a cluster of ChatGPT accounts that OpenAI assesses very likely originated in Russia and were used to generate social-media content promoting the International Burke Institute (IBI), a self-described expert community claiming a street address in Israel. It is the fifth report in the series OpenAI began in May 2024.
The report is a company self-report on activity conducted on its own platform. No independent verification of the account exists, and OpenAI's own attribution is expressly incomplete in one respect described below.
What OpenAI says it disrupted
The banned accounts prompted in Russian to generate social-media comments, most of them in English, posted on Substack, Telegram, X, Facebook and LinkedIn. OpenAI states that operators instructed ChatGPT to hide linguistic clues that they were Russian, and that because OpenAI does not permit access to its models from Russia the operators used VPNs.
Model-generated output fell into three lines of activity:
- Promotional posts. Social-media posts promoting articles on the IBI website, some from accounts bearing the IBI name and logo, others from accounts appearing to be everyday users whose activity consisted mainly of posting IBI articles.
- Replies on Substack. Replies to posts by real Substack users across a range of topics, typically including a request to follow the IBI channel, posted by the IBI account.
- German-language Telegram content and channel branding. One operator generated German-language posts for a Telegram channel called "Lahme Ente" ("lame duck") that criticized Ukraine, the EU and the German government and advocated better relations with Russia. A second operator generated logos for a dozen Telegram channels focused on Germany, the United States, France, Poland and Türkiye, and repeatedly requested Russian-language summaries of those channels' activity. One US-focused channel, the Telegram account "American Observer," posed as an American outlet with a bio OpenAI says contained multiple indicators of non-native language.
The AI / non-AI boundary
The report's central distinction is between what the models produced and what they did not. OpenAI states that the articles on the IBI website "were not generated from our models," and that the institute's country reports and its sovereignty index were likewise not model-generated. The activity OpenAI disrupted was the promotional layer surrounding the site, not the site's own content.
That non-AI core is described in detail. Of a sample of 36 articles linked to purported experts on the IBI site and published between September 2025 and May 2026, OpenAI states that 34 were copied from elsewhere on the internet, some years old and some attributed to the wrong authors. Two worked examples are given:
- An article on the China-Pakistan Economic Corridor, apparently copied from a Cambridge University Press original written by Professor Emeritus Yunas Samad of Bradford University and published on 7 July 2025, appeared on the IBI site attributed to Professor Katharine Adeney of the University of Nottingham and dated 28 December 2025.
- An article on migration governance, apparently copied from a Migration Policy Institute original published in April 2025 by Meghan Benton, Natalia Banulescu-Bogdan and Kate Hooper, appeared under an author page for "Kate Howell," dated 21 January 2026, using a photograph identical to a 2017 speaker page for an Australian professor of food chemistry of that name.
OpenAI reads the pattern as an attempt to make the site look legitimate by stocking it with authentic content while obscuring where that content came from. The site is stated to have been registered in February 2025 and, as of 17 July, to have claimed a range of world-class experts including Francis Fukuyama and Noam Chomsky. Some articles are described as apparently drafted by a Slavic speaker and machine-translated; the report gives as evidence the phrase "the Svetofor coalition" for Germany's traffic-light coalition, svetofor (светофор) being the word for traffic light in Russian and other Slavic languages.
The sovereignty index itself is presented as the operation's distinguishing feature. Country reports ranked and assessed states by the institute's own measure, focusing on countries critical of Russia's war on Ukraine — France, Germany, the European Union — with similar narratives about the United States. OpenAI reproduces samples, including a France report describing the country as "opened like a safe with historical capital, and now it is being sold off in parts," and a United States report stating that Trump "arrived [in Beijing] with the No. 1 sovereignty index in the world and left after taking a number of steps that [...] can be qualified as a voluntary reduction in autonomy."
Impact assessment
OpenAI assesses the operation's immediate impact as limited: typical posts received low view counts and official IBI accounts had low subscriber numbers, though the associated Telegram channels generally counted 10,000–20,000 followers each. Applying the Brookings Breakout Scale, OpenAI rates the operation at the lower end of Category Three — multiple platforms, with some indications of breakout to authentic audiences. The calibrated rating is the report's own; "limited impact" is OpenAI's characterization of audience reach, not of the scale rating.
The report locates the operation's significance in construction rather than reach: while ChatGPT produced only isolated promotional posts, those posts pointed to a credible-appearing institution with purported experts, republished academic work and a proprietary-looking risk index. OpenAI presents this as an illustration of AI used "as a supporting tool within a broader effort to manufacture authority, obscure the source of favored narratives, and establish assets that could be scaled over time," and observes that the supporting AI use is what exposed the broader operation.
OpenAI describes the campaign as new and previously unreported, and states that this is the first Russia-linked influence operation it has disrupted that went to such elaborate lengths, distinguishing it from others disrupted since the start of the war in Ukraine.
Limits the report states about itself
- Attribution to the Israeli side is incomplete. OpenAI states that online traces suggest a handful of real individuals in Israel may have represented the IBI in article submissions and at conferences, and that it is "not in a position to determine the relationship between these individuals, the IBI, and the operators in Russia."
- Origin is assessed, not established. The account cluster is described as one that "very likely originated in Russia."
- Scope is platform-bounded. The report covers activity OpenAI can observe on its own platform; the IBI website's own content falls outside what it disrupted.
Provenance
Published at openai.com on August 25, 2026, under OpenAI's Global Affairs category. The capture is complete in text; screenshots referenced in the report are represented by the descriptions in the body. The report links four prior OpenAI threat-intelligence reports, of May 2024, October 2024, June 2025, and one undated, hosted on the company's CDN. Operation, actor origin, platforms and impact assessment were corroborated independently by CNBC of the same date (Source: cnbc.com).
Relationships
- supports: AI and Misinformation — a documented case in which AI-generated content formed the promotional layer of a larger non-AI operation.
- instance-of: AI and Authoritarianism — a state-aligned narrative operation using commercial AI services.
- related: AI in Elections and Democratic Institutions, Synthetic Media / Deepfakes, AI and Democracy — the information-environment concepts the operation bears on.
- related: OpenAI — the report is also a record of the company's own enforcement and threat-intelligence practice.
- related: AI Content Provenance — the operation's core deception was misattributed authorship rather than synthetic generation.