OpenAI has recently banned a network of Russian-linked ChatGPT accounts after determining they were used in a coordinated influence campaign. These accounts employed VPNs to bypass regional restrictions and generated content for social media platforms, including Substack, Telegram, X, Facebook, and LinkedIn. The content was tied to a group calling itself the International Burke Institute (IBI), described as an “expert community” based in Israel. OpenAI found that while some material on IBI’s website appeared to be copied from academic sources—with false attributions in some cases—much of the content seemed designed to mask its Russian origin.
Network Structure, Tactics, and Reach
The campaign’s structure went beyond generating posts. The IBI website, registered in February 2025, claimed global expertise in economics, politics, sociology, and international affairs. Alongside republished or attributed academic work, the site introduced a “sovereignty index”—also called the Burke Index—evaluating countries across seven dimensions: political, economic, technological, information, cultural, cognitive, and military sovereignty. Russia ranked third in the index, following only the U.S. and China. The website’s commentary portrayed Russia as strong in the military, economy, science, and culture, while acknowledging challenges like demographic decline and innovation bottlenecks.
The campaign mainly produced content in English. Its operators instructed ChatGPT to strip out any linguistic signals that might reveal Russian origin. Some Telegram channels linked to the operation drew between 10,000 and 20,000 followers each. Though audience sizes were modest, OpenAI emphasized the importance of the underlying infrastructure: the ability to manufacture legitimacy via a shadowy institute, risk index, expert endorsements, and published academic texts distorted with misattribution and plagiarism.
The Role of AI
OpenAI’s investigation found that while most website articles didn’t appear to be AI-generated, some were machine-translated or produced by individuals with Slavic backgrounds. The use of AI centered primarily on content for social media posts—crafted to promote the IBI’s narratives—and for creating assets like profile images. One user asked ChatGPT to fashion a profile picture for a Telegram channel called “American Observer” and translated aspects of content to obscure origin.
The operation is considered distinct from previously known Russia-linked influence efforts largely because it created its own intellectual infrastructure: an institute-like entity, a proprietary sovereignty metric, and content masquerading as expert analysis. OpenAI warned that influence actors can use AI tools as modular elements of broader propaganda or disinformation machines, combining content generation with false authority frameworks to mislead audiences.
The operators appear never to have gained mass reach, but OpenAI stresses that the influence of an operation isn’t measured only in immediate audience size. The credibility built by an institute façade, the use of academic-styled reports and false attributions, and indices can lay groundwork for bigger campaigns.
OpenAI’s action—shutting down accounts and dismantling parts of this infrastructure—highlights rising concerns over AI-assisted disinformation. Key lessons: influence operations are increasingly masquerading their origin, using elements like provenance-masking and content that appears scholarly. Monitoring begets exposure.
What this means is that platforms, researchers, and regulators need to focus not just on content volume, but on provenance and legitimacy structures. As AI tools continue lowering entry barriers for producing polished, persuasive output, dark patterns of influence campaigns could become more common. Look ahead to new policies, detection tools, and perhaps legal frameworks designed to identify not just misleading content, but misleading institutions as well.
