OpenAI busts Iranian influence op using ChatGPT for election meddling

If you came across heated commentary online about the U.S. election recently, there is a chance the person behind the screen was not even a voter.

OpenAI announced that it banned a cluster of ChatGPT accounts tied to a covert Iranian influence operation known as Storm-2035. The group was leveraging ChatGPT to generate content aimed at swaying public opinion across several contentious topics, including the 2024 U.S. presidential campaign.

Playing both sides of the aisle

According to OpenAI, the Iranian network used ChatGPT for two main tasks: drafting long-form opinion pieces and generating short social media comments.

The long-form operation fed five pseudo-news sites posing as both progressive and conservative outlets, including domains like niothinker[.]com and savannahtime[.]com. Articles covered U.S. politics, candidates on both sides, the conflict in Gaza, and Israel’s presence at the Olympic Games, with occasional commentary on Venezuelan politics and Scottish independence.

On social platforms, the group operated a dozen accounts on X and one on Instagram, posting in English and Spanish. They frequently took real comments from other users and asked ChatGPT to rewrite them to blend in. Curiously, the operators also mixed in posts about fashion and beauty, presumably to look like authentic everyday users and slowly build a following.

Zero likes, zero traction

Despite the effort, the campaign fell almost completely flat. OpenAI found no evidence that the content reached a meaningful audience, noting that most posts received few to zero likes, shares, or comments.

On the Brookings Breakout Scale, which scores covert influence operations from 1 to 6, OpenAI placed this campaign at the low end of Category 2. That means content was scattered across platforms, but real people never picked it up. OpenAI credited threat intelligence published by Microsoft the week prior for helping uncover the activity before taking the accounts offline.

Why it matters

Generative AI has undeniably dropped the cost of spinning up a propaganda machine down to near zero. Anyone can now generate thousands of plausible political arguments across multiple languages and ideological stances in seconds.

However, this incident proves that generating content is not the same as generating influence. OpenAI’s public takedown cleverly reassures the public and regulators during an anxious election year, showing that foreign state actors are still struggling with distribution and genuine engagement.

Still, we think this is only round one. Attackers will keep iterating until they figure out how to bypass social media algorithms and blend in more convincingly. While takedowns like this are a win for platform safety, constantly playing whack-a-mole against zero-marginal-cost content generation remains an uphill battle.