OpenAI says it disrupted a covert influence campaign
OpenAI says it has disrupted a covert Russian influence campaign that used ChatGPT to generate social media content and conceal its operators’ origin. According to the supplied reporting, the network accessed the platform from Russia through VPNs and explicitly instructed ChatGPT to remove linguistic clues that might reveal Russian authorship. That detail matters because it shows generative AI being used not only for scale, but also for stylistic laundering: producing content meant to feel native to target audiences while hiding the source behind it.
The campaign’s apparent goal was to promote the “International Burke Institute,” or IBI, a purported think tank presented as being based in Israel. The organization pushed materials including a “sovereignty index” that ranked Russia favorably relative to Western countries. In practice, the operation appears to have mixed geopolitical messaging with the construction of a fabricated institutional front, giving propagandistic claims the appearance of research-backed analysis.
A familiar tactic with a more efficient toolchain
The operational pattern described here is recognizable from earlier information operations: create a seemingly independent outlet, publish content that flatters one side and discredits another, then distribute that material across multiple platforms until it begins to circulate on its own. What changes with AI is the efficiency of the pipeline. Instead of relying entirely on human writers or obvious copy-and-paste translations, operators can generate posts rapidly, adapt tone for different countries, and clean up language signals that might otherwise expose the campaign’s origin.
The reporting says the campaign spread content across X, LinkedIn, Facebook, Substack, and Telegram. Some material was in German, including posts to a Telegram channel called “Lahme Ente” that criticized Ukraine, the European Union, and the German federal government while arguing for closer ties with Russia. A second operator reportedly created logos for about a dozen Telegram channels focused on Germany, the United States, France, Poland, and Turkey, while also requesting Russian-language summaries of channel activity. That combination suggests a coordinated attempt to build durable media-like assets rather than just fire off isolated posts.
OpenAI’s account of the campaign also points to the use of false authority as a central tactic. According to the supplied text, 34 of 36 expert-linked IBI articles published between September 2025 and May 2026 were copied from other sources, sometimes with fake author credits. One example cited was a Cambridge University Press article falsely attributed to a professor at the University of Nottingham. Another piece from the Migration Policy Institute was reportedly credited to an Australian food chemistry professor. Those details indicate the operation was not merely pushing opinions. It was manufacturing credibility through plagiarism, misattribution, and institutional mimicry.
Low reach, but a scalable architecture
By raw audience numbers, the campaign does not appear to have achieved major breakout success. The reporting says individual posts received very few views and that official IBI accounts had low subscriber counts. Even so, Telegram channels linked to the network each reportedly reached between 10,000 and 20,000 followers. OpenAI rated the operation at category three out of six on the Brookings Breakout Scale, meaning it had spread across multiple platforms and shown early signs of reaching real users.
That middle-tier assessment is probably the most important part of the story. A campaign does not need viral traction to matter if it is building infrastructure that can later be scaled. Fake think tanks, cross-platform distribution accounts, branded Telegram channels, multilingual workflows, and AI-assisted editing are reusable assets. Once assembled, they lower the cost of future influence efforts. In that sense, a limited campaign today may function as a prototype for a much larger one tomorrow.
This is where the role of generative AI becomes more structurally significant. The supplied reporting does not show AI single-handedly transforming a fringe operation into a mass persuasion machine. Instead, it shows AI helping operators produce content, localize messaging, disguise stylistic fingerprints, and maintain a broader publishing network with less labor. That is a quieter but potentially more consequential shift. It turns influence work into something that is easier to sustain, expand, and customize across languages and regions.
Why the disclosure matters
OpenAI’s disclosure adds to a growing record of AI platforms identifying and removing coordinated misuse tied to state-linked or politically motivated campaigns. The article notes that the company has previously disrupted several Russian influence operations abusing ChatGPT. Each takedown is useful, but each also demonstrates the same underlying reality: mainstream generative tools are now part of the contested information environment, and platform enforcement is becoming one of the front lines in limiting that abuse.
For policymakers and researchers, the case highlights an uncomfortable asymmetry. Building a campaign like this is comparatively cheap. Detecting it requires platform telemetry, pattern analysis, and often cross-platform correlation that outside observers do not easily have. That means the public usually learns about these operations after the fact, when a company discloses what it found or when investigators connect the dots independently.
The broader lesson is that the danger is not just synthetic text flooding the internet. It is the coupling of AI-generated content with fake institutions, copied research, fabricated attribution, and distribution systems tailored to specific national audiences. That combination can make low-quality propaganda look procedurally legitimate long enough to win attention, seed narratives, or give sympathetic communities material they can redistribute organically.
OpenAI’s account suggests this particular campaign remained limited in visible reach. But the company’s warning that the infrastructure could have been scaled is the part worth watching. Influence operations do not need to succeed spectacularly at first to become important. They need only to learn what works, harden their process, and keep lowering the cost of persuasion attempts. On that measure, the latest disruption is less a closed chapter than a sign of what the next generation of AI-enabled information operations may look like.
This article is based on reporting by The Decoder. Read the original article.
Originally published on the-decoder.com