OpenAI has banned a new cluster of accounts linked to a recidivist Iranian influence operation, tracking it as STORM-2035. The group was caught using ChatGPT to churn out political comments in English and Spanish, which were then posted to X by a network of likely fake personas. It’s a direct sequel to an operation OpenAI first disrupted in August 2024, and the persistence shows that even after getting booted, state-linked actors will simply retool and try again.
The mechanics were clunky but deliberate. Operators prompted the model in Persian, requesting short comments—usually around a dozen at a time—on a scattered set of topics. The output covered Latino rights in the US, Scottish independence, Irish reunification, Venezuelan politics, Palestinian rights, and Iran’s military and diplomatic standing. Spanish-language tweets focused on Latin America and Iran’s geopolitical posturing, while English tweets targeted UK and Irish audiences. The accounts themselves were classic sock puppets: they followed far more people than followed them back, and many used profile pictures of young women with obscured faces, some lifted from Pinterest.
One notable shift from the group’s earlier playbook is a more disciplined approach to operational security. Last year’s version spread content across X, Instagram, and standalone websites. This time, the public-facing comments lived solely on X, suggesting an attempt to keep workstreams isolated—perhaps to make cross-platform takedowns harder or simply to test a narrower distribution channel. Either way, the effort flopped. The typical tweet racked up 150 to 350 views and generated precisely zero likes, shares, or comments. There’s no evidence any real users amplified the content.
That zero-engagement number is the real story here. These operations often sound ominous, but this one lands squarely in the lowest tier of the IO Impact Breakout Scale: activity on a single platform with no organic pickup. The gap between the operation’s ambition—commenting on everything from Scottish independence to Iran’s negotiating leverage—and its actual impact is almost comical. Still, the recidivism itself is worth noting. It confirms that low-cost, AI-assisted influence campaigns are now a permanent fixture of the information landscape, even if most of them are screaming into a void.
