OpenAI’s threat intelligence team has banned a cluster of accounts tied to an Iranian influence operation that was using its models to churn out propaganda. The group, linked to an entity called the International Union of Virtual Media, or IUVM, was using ChatGPT to generate and translate long-form articles praising Iran and the ‘Axis of Resistance’ while attacking the US and Israel. It’s a familiar playbook with a digital twist, but the execution here was remarkably shoddy.
This isn’t IUVM’s first rodeo. The open-source community has been tracking them since 2018, and the FBI seized their earlier domains back in 2020. This time around, the operation focused on producing content in English and French for their new website, iuvmpress.co. The automation was clumsy. OpenAI noted that on one occasion, the model’s own response message was accidentally published as a website tag—a mistake that screams either full automation with zero human oversight or a truly spectacular failure in proofreading. The articles were typically generated just a day before publication, and website tags were created immediately prior, pointing to a conveyor-belt approach to propaganda.
If the goal was influence, the results were a spectacular failure. Using the Breakout Scale, which rates information operations from 1 to 6, OpenAI pegged this as a Category 2: activity on multiple platforms but zero breakout and no meaningful audience. The numbers tell the story. Beyond the website, IUVM-branded accounts were found on TikTok, VKontakte, and Odnoklassniki, with follower counts of 10, 76, and 274, respectively. That’s not an influence network; that’s a group chat. The takedown by OpenAI is a minor speed bump for an operation already flattened by repeated social media bans and the earlier FBI domain seizure.
The broader lesson here isn’t about the sophistication of AI-powered propaganda—it’s about its persistent mediocrity in cases like this. While the industry fixates on deepfakes and hyper-personalized manipulation, state-linked actors are still failing to get more than a few hundred people to notice their work. The real value of these disclosures from OpenAI isn’t in stopping IUVM, which was already floundering. It’s in making the patterns of failure visible, showing that churning out AI-generated text isn’t a magic wand for relevance. If a propaganda network falls in a forest and no one is there to read it, does it make a sound? For IUVM, the answer is pretty clearly no.
