OpenAI announced on October 8, 2026, in a safety post that it had banned two clusters of accounts behind covert influence operations, one originating in Russia and one in Iran. Both used its models alongside conventional techniques to build seemingly independent "false front" entities and launder geopolitical, conflict-related messaging into target audiences. On OpenAI's breakout scale for influence operations, the Russian campaign was assessed at Category 5 — the highest level it has disrupted so far — and the Iranian one at Category 4.
The Iranian operation: seven fake journalists, roughly a hundred published articles
The Iranian operation maintained seven journalist personas that pitched long-form articles to small and medium online outlets around the world. According to CNN's same-day reporting, about a dozen outlets ran roughly one hundred of these articles; none were major mainstream publications, though some have sizable followings. The accounts prompted ChatGPT in Farsi to polish articles and pitches, and generated batches of social media comments on topics related to the US-Iran war. OpenAI said it could not identify the actor, describing activity consistent with a commercial, for-hire influence operator rather than a nameable government body.
The Russian operation: unwitting locals fronting a "think tank"
The Russian operation took a different route: it co-opted unwitting people in Latin America to run a self-styled research platform on the ground. Beyond operating that front, it fabricated "leaked" documents and audio scripts, some of which OpenAI confirmed were spread online. Both operations also leaned heavily on AI to draft internal reports, and used questionable or outright deceptive methods to inflate their own effectiveness — even the numbers reported back to their sponsors were padded.
Why fronts matter more than bot accounts
OpenAI says it has exposed 30 covert influence operations over the past two and a half years. Its judgment in this report: the operations with the highest potential reach are not the ones spinning fake social accounts, but the ones that land content in real media outlets. History supports the logic — earlier front-based operations built on invented journalist personas and fake newsrooms tended to stop once exposed, which is why OpenAI treats public naming as part of the remedy itself.
What newsrooms and readers can do
For editors, the lesson is concrete: verifying a contributor cannot stop at a résumé and a headshot; the provenance of long-form pitches and the background of the outlet behind them deserve one more check. For readers, one extra habit helps — look at which institution stands behind an article, and whether that institution has a verifiable address and track record. AI has lowered the cost of building a convincing front, but fronts still need human hands to spread; verification habits are currently the cheapest line of defense.
