OpenAI announced the identification and blocking of a coordinated group of ChatGPT accounts that, in its assessment, were operated by actors based in Russia. According to the disclosure, the network's participants used the neural network not for routine tasks but as a tool for an information operation: via VPN, they entered prompts in Russian and instructed the model to generate public comments in English and German, thereby erasing the linguistic markers of Russian origin. The generated content was then posted on X, LinkedIn, Facebook, Substack, and Telegram, and its central goal was to promote the website of the International Burke Institute (IBI) — an entity styled as a think tank with an address in Israel.
The Mechanics of the Operation: From Russian Prompt to 'Western' Comment
According to OpenAI, the scheme was organized as a conveyor. Operators, hiding behind virtual servers, formulated their tasks in Russian, and ChatGPT translated them into public texts in European languages. This approach allowed them to avoid the typical errors and stylistic traces by which Russian authors are usually identified. The resulting materials were distributed across several social platforms simultaneously, creating the illusion of independent Western commentators discussing the same topics. It was precisely this multi-platform presence and linguistic masking that, in the company's assessment, became the key signs by which the operation was singled out from general traffic.
Fake Authority: Plagiarism and False Authorship
The core of the entire construct was the IBI website, which, according to available data, was registered in February 2025 with a listed address in Israel. The platform claimed collaboration with leading world scholars, including Noam Chomsky and Francis Fukuyama. However, a content analysis revealed mass plagiarism: 34 of the 36 analyzed articles were copied in full from real academic sources, including Cambridge University Press and the Migration Policy Institute. Moreover, the works were attributed to unrelated scholars — in one case, an article on migration was signed with the name of an Australian professor of food chemistry who had no connection to the topic. This combination of borrowed text and other people's names created the appearance of a legitimate research institution.
Linguistic Traces and the Fabricated 'Sovereignty Index'
Despite the masking, artifacts of machine translation from Russian remained in the texts. A characteristic example cited by OpenAI was the rendering of the German term 'traffic-light coalition' in an English text as 'Svetofor coalition' — a literal calque of the Russian word 'svetofor' (traffic light), which gave away the source language of the prompt. The main ideological tool of the campaign was a fabricated 'sovereignty index': in reports using this scale, Russia was praised for its autonomy, while France, the United States, Germany, and the EU were subjected to harsh criticism. Thus, the formally 'independent' analytical product in fact reproduced Kremlin narratives, wrapped in academic style.
Contradictory Data
There is a notable gap between the public image of the IBI website and its actual indicators of influence. On the one hand, the platform positioned itself as an institute involving figures such as Chomsky and Fukuyama, which should have given it weight in the expert community. On the other hand, according to OpenAI's assessment, the direct reach of its social media publications was low, and the only confirmed results were Telegram channels that attracted between 10,000 and 20,000 subscribers each. The claimed collaboration with the named scholars is not corroborated by their participation and is in fact part of a constructed authority. Thus, the scale of the 'institute' in its self-presentation corresponds neither to the level of content originality nor to the actual audience figures, which in itself serves as an indicator of the structure's fabrication.
Assessment of Scale and a New Trend in AI Disinformation
On the Brookings Breakout Scale, which characterizes the maturity of information operations, this campaign received a category 3 rating — penetration across several platforms with isolated reach to real audiences. This means that the operation was technically deployed but did not achieve mass influence. Nevertheless, OpenAI emphasized that the case demonstrates a qualitatively new trend: the use of generative AI not only to create individual fakes but to construct an entire false authority — an imitation of legitimate think tanks featuring plagiarism, fake authors, and fabricated indices. For platforms and fact-checkers, this means the need for new methods of detecting not individual false claims but entire fabricated 'institutions' whose legitimacy rests exclusively on machine-generated content.
