OpenAI Takes Down ChatGPT Accounts Running Covert Iranian Influence Operation

OpenAI has banned a cluster of ChatGPT accounts that generated content for a covert Iranian influence operation known as Storm-2035, including commentary on candidates on both sides of the U.S. presidential election. The company disclosed the takedown this week and said it continues to monitor for further attempts to violate its policies.

The operation used ChatGPT to produce long-form articles and shorter social media comments, according to OpenAI. It published articles on U.S. politics and global events across five websites that posed as both progressive and conservative news outlets. A second workstream generated short comments in English and Spanish, posted through a dozen accounts on X and one on Instagram. Some of the X accounts posed as progressives, and others as conservatives.

OpenAI's investigation benefited from information about the operation published by Microsoft last week, which identified the Iranian effort as Storm-2035 and linked it to targeting of the 2024 U.S. election.

Low reach, but a serious signal

Despite the breadth of the operation's output, OpenAI found little evidence that anyone was paying attention. "This operation does not appear to have achieved meaningful audience engagement," the company said. The majority of social media posts identified by OpenAI received few or no likes, shares, or comments, and the company found no indications that the operation's web articles circulated on social media.

OpenAI assessed the operation using Brookings' Breakout Scale, which measures the impact of covert influence operations on a scale from 1 to 6. Storm-2035 landed at the low end of Category 2: activity on multiple platforms, but no evidence that real people picked up or widely shared the content. Category 6 represents the highest-impact operations on the scale.

The assessment echoes OpenAI's May report on state-affiliated threat actors, in which the company disrupted covert influence operations that similarly failed to gain traction.

How the operation worked

According to OpenAI, the operation used ChatGPT for two distinct purposes. The first workstream produced long-form articles on U.S. politics and global events, published on five websites that masqueraded as news outlets across the political spectrum — some posing as progressive, others as conservative. The second workstream created short comments in English and Spanish for social media.

The operation generated content on several topics. Its main focuses were the conflict in Gaza, Israel's presence at the Olympic Games, and the U.S. presidential election. To a lesser extent, it covered politics in Venezuela, the rights of Latinx communities in the U.S. — in both Spanish and English — and Scottish independence.

The fake personas also mixed their political output with lighter material. OpenAI said the accounts interspersed their political content with comments about fashion and beauty, "possibly to appear more authentic or in an attempt to build a following."

One technique stood out: some of the social media comments were produced by asking OpenAI's models to rewrite comments posted by other social media users — a low-effort amplification method that recycles authentic speech as covert propaganda.

Why it matters

The takedown comes amid a wave of elections in 2024, including the U.S. presidential race in November. OpenAI framed the case as part of its broader effort to prevent abuse and improve transparency around AI-generated content, including covert influence operations that "try to manipulate public opinion or influence political outcomes while hiding the true identity or intentions of the actors behind them."

The case also illustrates the emerging role of AI companies as frontline actors in election security. OpenAI said it has expanded this work throughout the year, in part by leveraging its own AI models to better detect and understand abuse — a defensive use of the same technology being weaponized by influence operators.

Foreign influence operations targeting elections are not new, but generative AI lowers the cost of producing plausible, voluminous content in multiple languages across fake personas. Storm-2035's failure to attract an audience in this instance does not guarantee that future operations will fare the same way, which is why OpenAI says it treats the activity as serious regardless of reach.

"Notwithstanding the lack of meaningful audience engagement resulting from this operation, we take seriously any efforts to use our services in foreign influence operations," the company said.

Sharing intelligence with stakeholders

Beyond removing the accounts, OpenAI has shared threat intelligence from the investigation with government, campaign, and industry stakeholders to support wider disruption of the activity.

OpenAI said it remains dedicated to uncovering and mitigating this type of abuse at scale by partnering with industry, civil society, and government, and by "harnessing the power of generative AI to be a force multiplier in our work." The company committed to continuing to publish findings like these to promote information-sharing and best practices across the ecosystem.

With the U.S. election approaching, the episode establishes a pattern that platforms, researchers, and campaigns should expect to recur: foreign operators experimenting with AI-generated propaganda, and AI vendors positioned as the first — and so far most effective — line of detection.