There is something uniquely unsettling about the way modern disinformation works. It no longer needs to be loud, crude, or obviously fake. Instead, it can be quiet, polished, and dressed up in the robes of academic respectability. That is precisely what OpenAI says it discovered when it disrupted a social media influence campaign believed to originate in Russia. The operation was not about crashing elections or spreading wild conspiracy theories. It was more patient and insidious: it sought to convince Western audiences that Russia was superior to its adversaries, that democratic nations were sacrificing their sovereignty, and that the countries supporting Ukraine were, in fact, the real problem. To do that, the operators behind the campaign built an entire “expert community” from the ground up—a fictional think tank called the International Burke Institute, or IBI. On paper, the IBI looked like a legitimate academic body. In reality, it was a Potemkin village of stolen research, borrowed reputations, and AI-generated content, all designed to give a pro-Russian narrative the weight of scholarly authority. The whole episode is a reminder that artificial intelligence tools, which can create text and images in seconds, are not inherently good or bad. They simply amplify what people point them at—and in this case, they were pointed firmly at the West’s vulnerabilities.
The architecture of the campaign is what makes it stand out from the usual run-of-the-mill bot networks. Rather than simply blasting out propaganda from anonymous accounts, the actors involved constructed a detailed fake infrastructure that could fool a casual observer. At the center was the International Burke Institute itself. The name sounds respectable, even distinguished, evoking the eighteenth-century philosopher Edmund Burke. But the reality was far less credible. According to OpenAI, the IBI’s website featured academic work that had been stolen from real authors and then attributed to entirely different people. In some cases, the original work was legitimate and well-researched, but it was presented in a way that made the site look more authentic, while also hiding the true origins of the material. The site was registered to an address in Israel, and there were signs that real individuals in Israel may have represented or promoted the page, presumably without understanding—or perhaps willingly ignoring—the darker purpose behind it. But the heart of the campaign was unmistakably oriented toward Moscow. The entire point was to persuade readers that Western nations were losing their autonomy, that their leaders were weak, and that Russia had the strength and wisdom to dominate in every important area of national power. In other words, it was a carefully curated mirror of reality meant to flatter Russia and denigrate the West.
For all the effort put into covering their tracks, the operators left behind a few telling slips. OpenAI noted that many of the prompts used to generate content included explicit instructions to hide any linguistic clues that might reveal the users were Russian speakers. That kind of care shows a level of operational discipline not often seen in social media influence campaigns. But language is stubborn. One of the more revealing mistakes involved a reference to Germany’s governing coalition, known colloquially as the “traffic light coalition” because of the colors of its member parties: red, yellow, and green. The IBI website referred to it as the “Svetofor coalition.” Svetofor is the Slavic word for traffic light. It is not a term a native English speaker would naturally use when describing German politics. It was a tiny crack in the facade, but a revealing one—evidence that the articles were likely drafted in a Slavic language and then translated, perhaps with the help of AI, into English. The campaign also made heavy use of ChatGPT to expand its reach. Some accounts promoted the IBI across X, LinkedIn, Facebook, Substack, and Telegram, generating not only captions but also AI imagery designed to make the posts look polished and credible. Other accounts were programmed to automatically engage with comments from real users on Substack, giving the fake think tank a semblance of ongoing conversation and organic interest. The operators asked ChatGPT for performance summaries of their pages—in Russian—and even used the tool to generate matching profile pictures for their fictitious network of commentators.
The centerpiece of the IBI’s intellectual facade was something called the Burke Sovereignty Index. According to the website, this index was designed to measure how well countries perform across a range of categories: political, economic, technological, informational, cultural, cognitive, and military. On its surface, it sounds like a reasonable academic project, the kind of comparative metric that policy analysts might cite in a journal article or a think tank white paper. But the results are so bizarre that they undermine any pretense of methodological seriousness. The Vatican City, for example, which covers less than half a square kilometer and has the smallest economy in the world, was ranked above Spain in the average of its scores. Russia, meanwhile, was given the highest possible military score—a claim that would be laughable to anyone who has followed the war in Ukraine, where the Russian military has repeatedly struggled against a smaller, less well-equipped opponent. The purpose of the Index, OpenAI concluded, was not to provide genuine insight. It was to make Western countries appear weaker, less stable, and less sovereign than Russia. By adjusting the rankings and weighting the factors in such a way that Russia consistently came out on top, the IBI could feed a narrative of Western decline and Russian resurgence. To a reader who did not look too closely, the charts and tables would appear authoritative. To a statistician, they were pure nonsense dressed up in bar graphs.
The IBI also published a list of expert contributors, and this is where the deception becomes truly brazen. The website included names like Francis Fukuyama and Noam Chomsky, both genuinely influential scholars whose work is widely known. It also listed former high-ranking US officials, including Mike Pompeo and Joseph Nye. You might wonder how such prominent figures ended up attached to a questionable think tank. They didn’t—at least not voluntarily. The affiliations were fabricated, and according to OpenAI, the site’s listing of these experts was not referenced anywhere in their actual biographies. Even more telling is the fact that some of the listed contributors had died years before the IBI was ever created. Shlomo Avineri and Jiang Ping were both included, despite being deceased at the time the site was active. That alone should have been a red flag, but it also shows how careless the operators could be when they thought no one was looking. To make matters worse, much of the academic work cited on the IBI website was real, but it was often stolen from its actual authors and misattributed to someone else entirely. This served two purposes: it made the website look credible because the sources were genuine, and it covered the tracks of those behind the operation, making it much harder to trace the intellectual currents back to their true origin. For the people whose work was hijacked, this is a violation of their name and their labor. For the public, it is a warning about the difficulty of trusting even polished, citation-heavy sources online.
What was the overall impact of this elaborate network? According to OpenAI, the reach was fairly limited, at least in terms of the numbers. Some of the Telegram channels involved managed to attract followings of between ten thousand and twenty thousand people, which is a meaningful audience but nowhere close to a global-scale influence operation. There is also little evidence that the IBI’s messaging broke through to mainstream audiences in a significant way. In some sense, this is the good news: even a sophisticated, AI-assisted disinformation campaign can fail to move the needle when the underlying message does not match people’s lived experience. But the takeaway should not be relief. The real lesson from this operation is the level of dedication and infrastructure that went into building something that looked so authentic. The site was designed to appear bona fide, using real academic research—even if mislabeled and misattributed—and presenting itself as a collection of experts. To the average person scrolling through social media, a brief visit to the IBI website would suggest legitimacy. That is precisely what makes these campaigns dangerous. As AI tools become more powerful and more accessible, the cost of creating convincing fake institutions will keep falling. The next operation might be even harder to detect, with better translations, more realistic imagery, and more carefully chosen real experts whose names can be borrowed. The answer is not to dismiss every unfamiliar source, but to cultivate a healthy skepticism, to ask who is behind the research, to check whether the experts actually endorse the organization claiming them, and to remember that even the most professional-looking website can be a machine built to deceive. In the battle between authenticity and manipulation, AI has made the playing field much more treacherous. But as this case shows, the architecture may be clever, the language may be polished, and the names may be impressive—yet the truth has a way of leaking through, sometimes in a single word like “Svetofor.”