OpenAI busts Russian and Iranian covert AI influence networks

Influence peddlers have officially moved past low-effort bot spam. Instead of flooding social feeds with repetitive replies, state-backed campaigns are now spinning up fake think tanks and fictional columnists to sneak geopolitical narratives into real news outlets.

OpenAI just revealed that it banned two state-sponsored covert operations originating from Russia and Iran. Both ran sophisticated false-front entities powered by ChatGPT to launder propaganda. The Russian campaign, codenamed "Dark Clark," reached a Category 5 on the IO Breakout Scale. It marks the first time OpenAI has encountered an operation with that level of reach across 30 influence campaigns tracked over the past two and a half years.

Fake think tanks and freelance journalists for hire

The Russian operation focused heavily on Latin America, running campaigns designed to smear Ukraine and stir local politics in Argentina, Bolivia, and Ecuador. The operators operated from Russia via VPNs and prompted mostly in Russian, setting up a faux research platform called the Social Research Center.

Instead of spamming AI junk, the group actually hired unsuspecting local Latin American researchers. A fictitious director named Mia Clark managed wages, hiring, and firing behind the scenes, effectively using duped staff to publish original work. The operators also claimed responsibility for planting viral hoaxes, such as claiming Argentine President Javier Milei bought Cartier collars for his dogs, and even paying locals to spray anti-Milei graffiti in Buenos Aires. Researchers have tied some of these tracks to Politology, an entity linked to the late Yevgeniy Prigozhin’s network.

Meanwhile, the Iranian operation, dubbed Bogus Bylines, crafted seven fictional journalist personas. Operators fed Persian instructions into ChatGPT to draft long-form opinion pieces, pitching them to regional outlets worldwide. One piece, carrying the fake byline Ervin Hoskins, was picked up by a local newspaper in Florida. OpenAI noted that the Iranian campaign appeared to be run by a commercial vendor operating as a for-hire influence outfit.

Ironically, both groups used ChatGPT far more to write internal progress reports to their handlers than to churn out public propaganda, frequently exaggerating their own metrics to claim credit.

Why it matters

This marks a tactical evolution in digital propaganda. As social networks get better at purging fake accounts, bad actors are realizing that penetrating real media publications delivers vastly superior reach and credibility.

Generative AI does not create new disinformation tactics out of thin air, but it dramatically lowers the cost of running long-term cover identities. A solo operator can now draft fluent, bespoke pitch emails and convincing long-form essays in multiple languages, making resource-strapped newsrooms prime targets.

We think OpenAI's decision to name and shame these networks is effective, because false fronts lose almost all their utility once their cover is blown. But the systemic risk remains high. As long as commercial influence firms can access frontier models through VPNs and layered accounts, media gatekeepers will face an uphill battle verifying who is actually behind the pitch.