OpenAI has banned a cluster of ChatGPT accounts that it says very likely originated in Russia and were used as part of a covert influence campaign built around the International Burke Institute (IBI), a self-described “expert community” claiming to be based in Israel. According to OpenAI, the operators used ChatGPT primarily to generate social media posts promoting the organization across X, LinkedIn, Facebook, Substack and Telegram, while attempting to conceal their Russian origins.
OpenAI said the operation was more elaborate than other Russia-linked influence campaigns it has disrupted since the start of the war in Ukraine, even though its direct audience appears to have been relatively small. The broader operation involved a website containing copied and misattributed academic work, purported experts and a “sovereignty index” that portrayed Russia favorably while criticizing Western countries. OpenAI said the website content and sovereignty reports were not generated using its models; ChatGPT was primarily used to produce promotional social media content.
OpenAI Identifies Russia-Linked ChatGPT Accounts
The investigation began after OpenAI identified AI-generated social media content associated with the International Burke Institute. The company said the accounts involved were very likely operated from Russia and used ChatGPT to generate posts, mostly in English, despite the operators prompting the model in Russian.
OpenAI said the operators specifically instructed ChatGPT to remove linguistic characteristics that could reveal their Russian background. Because OpenAI does not provide access to its models from Russia, the operators allegedly used VPNs to connect to the service.
How The AI Was Used
| Activity | Reported Use Of ChatGPT |
|---|---|
| Social media posts | Generated promotional content for IBI |
| Language | Operators prompted in Russian; much output was in English |
| Concealment | Instructions attempted to hide Russian linguistic clues |
| Platform targets | X, LinkedIn, Facebook, Substack and Telegram |
| Telegram content | German-language political posts |
| Channel development | Generated logos and requested channel summaries |
| Access method | VPNs used to circumvent Russia access restrictions |
The activity demonstrates that the AI component was primarily an amplification and content-production tool rather than the core source of the campaign’s narratives.
International Burke Institute Presented Itself As An Israeli Organization
The central entity in the campaign was the International Burke Institute, or IBI. Its website described the organization as an expert community based in Israel and presented an apparent network of prominent researchers and academics.
OpenAI said the website domain was registered in February 2025. As of July 17, the site claimed to feature experts including Francis Fukuyama and Noam Chomsky.
However, OpenAI’s review found substantial evidence that the site’s research material had been copied from other sources and, in some cases, attributed to the wrong authors.
OpenAI’s Review Of IBI Articles
OpenAI reviewed a sample of 36 articles linked to experts on the IBI website and published between September 2025 and May 2026.
| Finding | Number |
|---|---|
| Articles reviewed | 36 |
| Articles copied from elsewhere | 34 |
| Share copied | ~94% |
| Publication period examined | September 2025-May 2026 |
The company said some articles were years old, while others had been assigned to incorrect authors. One article concerning the China-Pakistan Economic Corridor appeared to have originated with Cambridge University Press but was attributed to a different academic. Another article on migration governance was reportedly copied from the Migration Policy Institute and attributed to an unrelated Australian professor.
These findings suggest that the website was designed to create an appearance of institutional credibility by combining authentic academic material with misleading attribution.
A “Sovereignty Index” Favored Russia
One of the most distinctive elements of the operation was the IBI’s so-called “sovereignty index.”
The index ranked countries according to a measure of national sovereignty and was used to portray Russia in a favorable light while criticizing Western countries and institutions. OpenAI said the index represented the first time it had disrupted a Russia-linked influence operation that went to such lengths to construct an apparent independent institution around a favorable geopolitical narrative.
Reports published around the index focused on countries including France, Germany and the United States, as well as the European Union.
Narratives Promoted By The Website
The reports generally criticized Western governments and institutions and argued for greater national autonomy or sovereignty.
OpenAI said the operation’s content included criticism of France’s government, skepticism toward U.S. policy and arguments that Germany needed greater military and economic independence from the United States.
The company emphasized that these sovereignty reports were not generated using ChatGPT. This distinction is important because the AI model was used to promote the material, rather than create the underlying reports themselves.
ChatGPT Content Was Distributed Across Multiple Platforms
The operators used ChatGPT-generated material across several major social platforms.
Some posts appeared on accounts carrying the IBI name and logo, while others were posted by accounts that appeared to be ordinary users. OpenAI said some of these accounts were likely inauthentic because their primary activity consisted of sharing IBI material.
The operators also used ChatGPT to generate replies to posts by genuine Substack users. Those replies generally promoted IBI and encouraged users to follow the organization’s channel.
Platform Distribution
| Platform | Reported Activity |
|---|---|
| X | IBI promotional posts |
| AI-generated promotional content | |
| IBI-related posts | |
| Substack | Promotional posts and replies |
| Telegram | Political content and IBI promotion |
The multi-platform strategy was intended to make the organization appear more established and potentially expand its reach beyond a single social network.
German-Language Content Targeted European Audiences
The campaign was not limited to English-language material.
OpenAI identified German-language posts generated by one operator for a Telegram channel called “Lahme Ente,” meaning “lame duck.” The posts criticized Ukraine, the European Union and the German government while advocating closer relations with Russia.
A second operator reportedly used ChatGPT to generate logos for around a dozen Telegram channels focused on Germany, the United States, France, Poland and Türkiye.
That operator also repeatedly asked ChatGPT for Russian-language summaries of activity on these channels.
One U.S.-focused channel, called “American Observer,” presented itself as an American outlet but contained indicators that its operators were not native English speakers.
The Campaign’s Reach Was Limited
Despite the elaborate infrastructure, OpenAI said the operation did not achieve substantial immediate reach.
Typical social media posts received relatively few views, while official IBI accounts had low subscriber numbers. The associated Telegram channels performed better, generally attracting between 10,000 and 20,000 followers each.
Reported Audience Impact
| Measure | Reported Result |
|---|---|
| Typical social media posts | Low view counts |
| Official IBI accounts | Low subscriber numbers |
| Telegram channels | Generally 10,000-20,000 followers |
| Overall impact | Limited |
| OpenAI assessment | Lower end of Category Three on Brookings Breakout Scale |
OpenAI said the operation fell at the lower end of Category Three on the Brookings Breakout Scale, which considers factors including the number of platforms involved and evidence that content reached authentic audiences.
The relatively modest reach means the immediate impact was limited, but OpenAI argues that audience size does not fully capture the significance of the operation.
Why The Infrastructure Matters
OpenAI said the most important feature of the campaign was the infrastructure created around the apparent think tank.
Rather than simply generating political comments, the operators built a website designed to look like a legitimate research organization. The site included purported experts, academic material and a proprietary-looking index.
The AI-generated social media posts then served as a distribution mechanism for that material.
This created a multi-layered system:
Russian-Based Operators
↓
ChatGPT Content
↓
Social Media Promotion
↓
International Burke Institute
↓
"Expert" Website + Academic Content
↓
"Sovereignty Index"
↓
Pro-Russia / Anti-West Narratives
The structure demonstrates how AI can be used as one component of a larger influence operation rather than as the sole mechanism for creating deceptive content.
OpenAI Says Some Israeli Individuals May Have Been Involved
OpenAI also identified online traces suggesting that a small number of real individuals in Israel may have represented the International Burke Institute in article submissions and conferences.
However, the company said it could not determine the relationship between those individuals, IBI and the Russian operators behind the ChatGPT activity.
That qualification is significant because it means OpenAI has not publicly established that every individual associated with the IBI brand was part of the Russian influence operation.
The company’s findings primarily concern the cluster of ChatGPT accounts and the associated online activity it was able to identify.
The Latest Action Fits A Broader Pattern
The operation is the latest in a series of Russia-linked influence campaigns that OpenAI says it has disrupted.
In previous reports, the company has described operations using its models to generate multilingual political comments, website content, translations and social media material. In 2024, OpenAI disclosed the Russia-origin “Doppelganger” operation, while another campaign dubbed “Bad Grammar” used AI-generated Telegram comments targeting audiences in Russia, Ukraine, Moldova, the Baltic states and the United States.
In 2025, OpenAI also reported disrupting “Helgoland Bite,” a Russia-linked operation that used AI to generate German-language political content about Ukraine, NATO and German domestic issues.
The latest operation stands out because of the effort to establish a seemingly independent research institution around the narratives being promoted.
The Bigger Picture
OpenAI’s latest enforcement action illustrates how generative AI is becoming an operational tool for influence campaigns rather than necessarily being the source of their underlying narratives. In this case, ChatGPT was used to create and distribute promotional content, while the wider operation relied on a website, purported experts, copied academic research and a proprietary-looking sovereignty index to manufacture credibility.
The limited audience reach shows that AI-assisted influence operations do not automatically translate into mass influence. However, the infrastructure built around IBI could potentially have been expanded over time. OpenAI’s decision to ban the accounts therefore reflects a broader effort to disrupt such operations before they can develop larger and more credible audiences.
Looking Ahead
OpenAI is likely to continue combining account enforcement with investigations of how its models are being used across external platforms. The company has increasingly published case studies describing the tactics it identifies, partly to help researchers, platforms and policymakers understand how AI is being incorporated into covert influence activity. The latest case also shows why monitoring the surrounding infrastructure can be as important as detecting individual AI-generated posts.
For the wider AI industry, the episode underscores the difficulty of preventing misuse when models can be incorporated into larger information operations. AI tools can accelerate translation, content production, branding and social-media distribution, even when the underlying narratives and websites are created separately. As a result, platforms will face continued pressure to identify coordinated deceptive behavior while distinguishing legitimate political activity, research and journalism from covert attempts to manufacture public credibility.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
