AI Pulse by Inblix

OpenAI Banned Iranian Influence Op STORM-2035—Its 12+ Fake X Personas Flopped

OpenAI Blog · Oct 1, 2024 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: OpenAI Banned Iranian Influence Op STORM-2035—Its 12+ Fake X Personas Flopped

OpenAI has pulled the plug on a cluster of ChatGPT accounts tied to an Iranian influence operation it’s calling STORM-2035, catching them red-handed generating polarizing content about the U.S. election. The actors weren’t subtle—they used the AI to churn out long-form articles for five phony news sites and short, punchy comments for over a dozen fake social media personas.

The operation ran two parallel tracks. One pumped English-language articles, some as long as 900 words, onto websites that cosplayed as both progressive and conservative outlets. The other workstream weaponized brevity, using ChatGPT to spit out comments on X and Instagram. An account on Instagram pretended to be a Scottish independence supporter, while more than a dozen X personas masqueraded as partisans for both major U.S. presidential candidates. It’s a classic “both sides” approach to sew division, but with a new AI-powered assembly line.

Microsoft had previously flagged some of these websites under the Storm-2035 moniker. Crucially, OpenAI says its unique visibility into model usage let it connect the social media posts and the websites to the same actor for the first time. Meta later confirmed the Instagram account was linked to an Iranian network it had disrupted way back in December 2022. The operators even had the audacity to ask ChatGPT to write an article claiming that Microsoft’s exposure of their operation proved Iranian cyber ops were more impactful than Russia’s or China’s. That request didn’t appear to get published.

Here’s the kicker: the entire campaign was a dud. OpenAI states the operation “does not appear to have achieved meaningful audience engagement.” Despite larding their feeds with non-political fluff about beauty and fashion to seem authentic, the fake personas couldn’t buy a real follower. It’s a stark reminder that while generative AI lowers the cost of producing disinformation, it can’t magically solve the hard part—getting anyone to actually care.

💡 Key Takeaways

  1. OpenAI connected fake social media accounts and phony news websites to a single Iranian influence campaign for the first time, using its visibility into how ChatGPT was being used.
  2. The operation attempted a 'both sides' playbook by creating fake personas that posed as supporters of both major U.S. presidential candidates, aiming to polarize voters.
  3. Despite using AI to scale content production, the operation was a complete failure in terms of reach, failing to generate any meaningful audience engagement.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

← Back to all articles