OpenAI busts Iranian op using ChatGPT for election spin

OpenAI has banned a cluster of ChatGPT accounts linked to an Iranian influence operation that churned out AI-generated content about the U.S. presidential election, the war in Gaza, and other divisive topics. The company identified the operation as Storm-2035, a nod to Microsoft’s earlier threat intelligence, and moved to take it down this week. The scheme’s output spanned five websites posing as both progressive and conservative news outlets, alongside a dozen accounts on X and one on Instagram. The accounts didn’t just compose original posts—they asked ChatGPT to rewrite comments from other social media users, a tactic designed to blend into existing political conversations.

Here’s the thing: the operation was a dud. By every available metric, it failed to gain traction. The majority of the social media posts OpenAI identified received few or no likes, shares, or comments. The company didn’t find any evidence that the web articles were shared across social platforms, either. Using the Brookings Institution’s Breakout Scale, which rates covert influence operations on a scale of 1 to 6, this one barely registered at the low end of Category 2. That means it had activity on multiple platforms, but no indication that real people picked up or widely shared the material. It’s the digital equivalent of shouting into a void.

What makes this operation interesting isn’t its impact—it’s the sheer breadth of the content farm. The actors didn’t just pump out straightforward political propaganda. They peppered their feeds with commentary on fashion and beauty, a likely attempt to appear more authentic or build a follower base before pivoting to heavier topics. The political content itself was all over the map, touching on the U.S. election from both conservative and progressive angles, Israel’s participation in the Olympic Games, Latinx community rights, Venezuelan politics, and even Scottish independence. That scattergun approach suggests a group casting a wide net, hoping something, anything, would stick.

The takedown is part of a broader push by OpenAI to use its own models to detect and understand abuse at scale, especially during a year packed with global elections. The company is sharing threat intelligence with government, campaign, and industry partners. Still, the low engagement numbers raise a question that hangs over a lot of these disclosures: if the content reached virtually no one, is the public really safer knowing the details, or does the announcement itself risk amplifying a narrative that otherwise would have fizzled out on its own?