OpenAI says it has disrupted a covert Iranian operation that used ChatGPT to produce political and news-style content for websites and social media. The accounts were linked to an influence campaign called Storm-2035, which generated material about the US presidential race and several other politically charged subjects.
The company suspended multiple ChatGPT accounts connected to the campaign. Microsoft also reported this week on AI-powered manipulation attempts from Iran and other countries, placing the activity inside a broader pattern of attempted online influence operations using generative AI.
What OpenAI says Storm-2035 did
According to OpenAI, the operation used ChatGPT to create content on several topics, including comments about candidates from both parties in the US presidential election. That material was then distributed through social media accounts and websites.
The campaign had two main content formats. One was longer-form article writing. The other was shorter commentary designed for social platforms.
In the longer-form approach, the operation created articles about U.S. politics and world events. Those articles appeared on five websites that presented themselves as progressive and conservative news outlets.
The shorter-form approach focused on brief posts in English and Spanish. OpenAI identified 12 accounts on X and one on Instagram involved in the operation. Some X accounts posed as progressive, while others posed as conservative.
The campaign covered divisive topics
The AI-generated content centered mainly on the Gaza conflict, Israel at the Olympics, and the US presidential election. To a lesser extent, it also covered Venezuelan politics, Latinx rights in the US in both Spanish and English, and Scottish independence.
That mix matters because the operation did not appear to be limited to one ideological identity or one issue. Based on OpenAI's description, the accounts tried to enter different political spaces by adopting both progressive and conservative personas.
The source does not say that the campaign successfully changed public opinion. It says the opposite: OpenAI reported that the operation failed to gain much traction.
- Most of the social media posts OpenAI found received few or no likes, shares, or comments.
- The articles showed no signs of being shared on social media.
- The campaign still shows how ChatGPT can be used to produce political content at different lengths and in different languages.
Why the low traction still matters
The limited reach is an important part of the story. Like covert campaigns reported in May, this operation appears to have produced content without producing much visible engagement.
That does not make the use of generative AI irrelevant. The case shows that a campaign can use ChatGPT to draft longer articles, generate social media commentary, and support accounts that present themselves as belonging to different political camps.
In practical terms, AI can reduce the effort needed to create posts and articles. The source describes the use of ChatGPT for both website content and social commentary, which means the same campaign could fill multiple channels with material that looks suited to each format.
Still, content volume is not the same as influence. OpenAI's findings, as described in the source, point to a campaign that created and published material but did not show strong signs of audience uptake.
AI misinformation is not only about fake content
The source also points to a broader issue: generative AI can be used in political propaganda in several ways, including mass-producing manipulative text and creating believable fake voices and images.
But the simplest deceptive use may be different. The mere existence of generative AI can make it easier for people to claim that real content is fake.
The source cites US presidential candidate Donald Trump as an example. He recently alleged that photos of an enthusiastic crowd at a Kamala Harris event were "A.I.'d", meaning manipulated with AI. The source says this is at least the second time that Trump has used this tactic.
That point connects to a 2017 prediction by deepfake technology creator Ian Goodfellow: people should no longer trust images and videos online. The concern is not only that false images, voices or videos can be made. It is also that the possibility of fakery can weaken trust in real audiovisual media.
The bigger lesson from OpenAI's suspension
The suspension of ChatGPT accounts tied to Storm-2035 shows both the capability and the current limits of this kind of operation. ChatGPT was used to create political articles and social posts, and the campaign tried to distribute them through websites, X, and Instagram.
At the same time, OpenAI said the material it found attracted little or no engagement in many cases. The websites' articles also showed no signs of being shared on social media.
For readers, the takeaway is direct: AI-generated influence attempts can be active without being effective. The harder problem is that the presence of AI now affects the whole information environment, from covert content production to public claims that authentic-looking media may have been manipulated.