JA EN
shiichan

Busted! OpenAI Takes Down Iran-Linked Influence Operation "STORM-2035"

OpenAI News ★★★ View Markdown
#openai#ai-safety#security

Hi, it's Shiichan! Today's story is less about excitement and more about relief, so let me walk you through an OpenAI report on countering misuse of AI.

OpenAI News openai.com

What was announced?

OpenAI's News reported that it banned ChatGPT accounts linked to STORM-2035, an Iran-based influence operation. The network used AI to produce political content about the US presidential election and UK-related topics, then spread it across several websites and social platforms.

Why it matters

Generative AI is really good at writing, which also makes it an easy target for misuse in influence operations. This takedown landed close to the 2024 US presidential election, so it's drawing attention as an example of countering election-related disinformation. Whether AI providers can actually catch and stop this kind of misuse matters for trust in AI as a whole.

What changes

By banning the accounts tied to this network, OpenAI cut off STORM-2035's ability to keep generating new content with ChatGPT. OpenAI also shared information about the operation with Microsoft and Meta. Meta confirmed that the network's Instagram account was connected to an Iranian network it had disrupted back in December 2022. It's a good example of platforms cross-checking notes to identify the same bad actors operating across different services.

Dive Deep

STORM-2035's output fell into two main buckets:

  • Long-form articles of roughly 700-900 words in English, posted to websites posing as either progressive or conservative outlets
  • Short comments in English and Spanish, meant to be spread across social media

The content showed up across 5 websites, plus more than a dozen fake accounts on X (formerly Twitter) and one on Instagram. The X accounts impersonated supporters of various US presidential candidates, while the Instagram account posed as a Scottish independence supporter. The topics ranged from the conflict in Gaza and Israel's presence at the Olympics to the US presidential election, along with Venezuelan politics and Latinx rights issues.

But the operation doesn't seem to have actually worked: engagement on the posts was low to none, and there was no sign the articles got meaningfully shared on social media. Some of the websites had already been identified by Microsoft under the name Storm-2035, and OpenAI's investigation built on that earlier work.

Wrap-up

  • OpenAI banned ChatGPT accounts linked to the Iran-based influence network STORM-2035
  • It churned out 700-900 word articles for sites posing as progressive or conservative outlets, plus short comments in English and Spanish
  • Content spread across 5 websites, a dozen-plus fake X accounts, and one Instagram account
  • Topics spanned Gaza, the Olympics, and the US election, plus Venezuelan politics and Latinx rights
  • Engagement was low and there was no real spread; sharing info with Microsoft and Meta helped confirm the network's identity

If you care about AI safety and countering election-related disinformation, this report is worth a read!

Related posts