1. Behind the sterile language of a corporate security announcement lies a story that feels almost like a cold-war spy novel, except the spy is not a person but a machine. This week, OpenAI revealed that it had banned a group of ChatGPT accounts traced to Russia, accounts that were being used to run a covert influence campaign designed to manipulate public opinion across the West. The operators of those accounts did not simply sit in Moscow and type angry posts; they built an entire fictional world. They invented a think tank, populated it with plagiarized academic articles, gave it a fake address in Israel, and used artificial intelligence to generate a steady stream of social media content that praised Russia while tearing down Western democracies. To bypass OpenAI’s restrictions on accessing its models from Russia, they disguised their internet connections with virtual private networks. They worked in Russian, but they asked ChatGPT to produce content in English, German, and other languages, instructing the model to scrub away any linguistic fingerprints that might betray a Russian origin. The content then found its way onto Substack, Telegram, X, Facebook, and LinkedIn. It was a sophisticated operation, not because it was loud or viral, but because it was patient, structural, and designed to look like a legitimate voice in the global conversation. In many ways, this is the new face of information warfare: not a troll farm screaming from the shadows, but a quiet, machine-assisted effort to manufacture credibility, one article, one post, one carefully worded data point at a time.
2. At the heart of this campaign was an organization called the International Burke Institute, a name that evokes serious, traditional scholarship and the kind of think tank that policymakers might actually cite. The group described itself as an expert community, and its website listed a street address in Israel, presumably to lend it an air of international legitimacy and gravitas. But when OpenAI’s investigators took a closer look, they found that the institute was largely a fabrication. The website, registered in February 2025, falsely claimed to feature prominent academics, including figures as well-known as Francis Fukuyama and Noam Chomsky, names that would instantly give an organization intellectual weight. Yet the articles that the institute linked to were not original scholarship. In a review of thirty-six articles attributed to the institute’s “experts,” published between September 2025 and May 2026, thirty-four were copied from other places on the internet, often with incorrect attribution. The entire enterprise was a collage of stolen words, repackaged under a fictional banner, designed to make readers believe that independent thinkers were reaching thoughtful conclusions about world affairs. The most striking piece of this fabricated landscape was a supposed “sovereignty index,” a data-driven ranking of countries that promised to measure national autonomy in a neutral, academic way. In reality, the index was rigged. It was built to cast Russia in a favorable light and to criticize Western nations, especially the United States, Germany, France, and the European Union. One entry about the United States described a visit by President Trump to Beijing, claiming that he arrived with the world’s top sovereignty index and left after taking steps that could be described as a “voluntary reduction in autonomy.” It was a deliberately humiliating narrative, dressed up in the cold, respectable vocabulary of political science, and it was being laundered through a fake institute with a stolen academic roster.
3. The campaign was not limited to English-language content. One of the operators, working in German, generated posts for a Telegram channel called “Lahme Ente,” which translates to “lame duck.” That channel criticized Ukraine, the European Union, and the German government, while pushing a narrative that called for closer ties with Russia. This is a common thread in Kremlin-aligned messaging: undermine the institutions that oppose Russia, sow doubt in Western leadership, and present Moscow as a reasonable, reliable partner. What made this operation different was the use of ChatGPT to produce those posts, and to do so at scale, with ease, and without leaving the telltale signs of non-native writing that often exposed past Russian trolling efforts. Another operator took on a different role, creating logos for roughly a dozen Telegram channels focused on Germany, the United States, France, Poland, and Turkey, and then asking for Russian-language summaries of how those channels were performing. The operation was, in effect, building a media empire of small, targeted channels, each with its own visual identity, each seemingly operated by locals, each pushing a slightly different angle tailored to its audience. This is the real craft of modern influence operations: not blasting the same message to everyone, but segmenting the audience, creating local voices, and cultivating specific grievances. The operators were not just generating text; they were creating an ecosystem of voices, a chorus of seemingly independent critics all singing from the same Russian hymn sheet. And they did it with the help of an AI assistant that followed their instructions to remove any trace of its own involvement, effectively helping them launder their propagandistic intentions into clean, plausible prose.
4. But here is where the story takes a strange and important turn. For all the effort, the reach of this operation was remarkably small. Most of the social media posts generated by the campaign received low view counts. The official accounts of the fake International Burke Institute had few subscribers. Even the Telegram channels associated with the campaign, which were probably the most successful element, attracted only between ten and twenty thousand followers each. By the standards of viral disinformation campaigns, which can reach millions through algorithmic amplification and media pickup, this operation was a failure. It did not move markets, elect candidates, or dramatically shift public opinion. It barely rippled across the surface of the internet. Yet OpenAI argued that the important thing here was not the actual reach, but the underlying infrastructure that the operators had constructed. The posts were designed to steer readers toward the fake research institute, to create the impression that a legitimate expert body was backing certain narratives. The goal was not necessarily to win the global argument today, but to build a foundation that could be expanded later, a reservoir of fake credibility that future operations could draw upon. In a world where a single viral moment can be engineered, the operators were playing a longer game. They were betting that if they could make their lies look respectable enough, with fake experts, fake indexes, and fake institutions, then eventually journalists, politicians, and the public would start treating them as a real source. This strategic patience is exactly what makes AI-assisted influence operations so dangerous, and why security researchers are increasingly concerned not about the obvious operations, but about the quiet, unremarkable ones that are laying the groundwork for something much bigger.
5. The broader implications of this operation are deeply unsettling, because they reveal how easily artificial intelligence can be weaponized to manufacture institutional credibility. In the past, running a fake think tank required significant resources. You had to hire people to write articles, design websites, manage social media accounts, and maintain a consistent public voice. You had to be fluent in the languages of your target audiences, and you had to avoid the mistakes that would expose your true origin. That was expensive, difficult, and risky. ChatGPT and similar models have changed the equation. Now, a small team of operators can generate thousands of words in dozens of languages, create logos, write tweets, draft Telegram posts, and even produce academic-sounding reports, all with a single AI assistant. They can tell the model to erase its own traces, to sound like a native speaker, to avoid any phrasing that might raise suspicions. The result is a dramatic lowering of the barrier to entry for influence operations. It means that the disinformation campaigns that once required state sponsorship and vast troll armies can now be run by a handful of people with a stolen database, some VPNs, and an internet connection. And when these operations are exposed, as they inevitably are, the illusion collapses, but only after it has served its purpose. The articles remain on the internet, the fake index has been cited somewhere, the seed of doubt has been planted. Even when the truth comes out, it often doesn’t entirely undo the damage. People remember the headline, the catchy statistic, the claim that some country has degraded its sovereignty, even if they later learn that the source was a fabrication. AI has given manipulators a powerful new tool for building these enduring half-truths, and this operation is likely just an early example of a trend that will only grow more sophisticated over time.
6. So what should we take away from this strange, almost absurd story of a fake think tank, a rigged sovereignty index, and a dozen small Telegram channels? The first lesson is that the problem of AI-generated disinformation is not hypothetical or futuristic; it is happening right now, and it is being produced by real people with real political agendas. The second lesson is that visibility is not the only measure of danger. This operation had almost no reach, but that does not make it irrelevant. It was a test, a prototype, a proof of concept. The operators were learning what worked, refining their methods, and building structures that could be used in future, larger campaigns. As AI models become more capable, these operations will become harder to detect, harder to trace, and harder to explain to a public that is already exhausted by endless online manipulation. The response cannot be left to AI companies alone, although their vigilance is important. Social media platforms, journalists, and ordinary users all need to be more skeptical of the sources that shape our political conversations. We need to ask who is behind an institution, who funds it, who its experts are, and whether its data can be independently verified. We need to understand that a polished website and a list of famous names can be manufactured in a matter of hours, and that the quiet, boring-looking think tank with a few thousand followers might be a seed being planted for something far more consequential. And above all, we need to remember that behind every machine-generated post, there is a human intention. There are people sitting in rooms, instructing an AI to rewrite the world in their image, to undermine our trust in each other and in the institutions that hold our societies together. The defense against that is not just technology, but awareness, community, and a stubborn commitment to the truth, even when it is messy, complex, and inconvenient. In the end, the story of the International Burke Institute should be a reminder that the internet is not a neutral space. It is a battlefield, and the weapons are words, generated at the speed of thought, aimed at the foundations of our shared reality.