OpenAI Shuts Down Iran-Linked Influence Network IUVM!

Hey, it's Shiichan! I found a slightly more serious piece of news today, a look behind the curtain at how AI gets misused, so let me walk you through it.

OpenAI News openai.com

What was announced?

OpenAI's News page announced that it had banned accounts linked to IUVM, an Iran-based influence network. IUVM used AI models to mass-produce pro-Iran, anti-Israel, and anti-US content aimed at English- and French-speaking audiences.

Why it matters

IUVM isn't a new name. The FBI actually seized its domains back in 2020. What OpenAI caught this time was activity tied to what looks like a successor site, iuvmpress.co. This matters because generative AI now lets operations like this automate article writing, translation, and even headline creation. Work that used to take a team of humans grinding out propaganda by hand can now be scaled up with AI, and that risk is no longer hypothetical.

What changes

OpenAI banned the accounts tied to this network, cutting off its pipeline for generating, proofreading, and translating articles with AI models. That said, OpenAI's own assessment placed the network at Category 2 on its Breakout Scale, meaning it posted across multiple platforms but never broke out or gained meaningful engagement anywhere. So this is a real example of AI-assisted influence operations, but not one that went viral.

Dive Deep

Here's what OpenAI observed about how IUVM actually worked:

• It used OpenAI's models to generate and proofread article bodies, headlines, and website metadata tags
• Content was published on iuvmpress.co (believed to be a successor to the domain the FBI seized in 2020) and iuvmarchive.org
• Most articles went live within a day of being generated, and tags were often created right before posting, a pattern that points to automation
• In at least one case, leftover text from an OpenAI model's response accidentally ended up published in a site's tag field
• The messaging consistently praised the Palestinian cause and the so-called "Axis of Resistance" while pushing anti-US and anti-Israel narratives

That leftover model-response text sitting in a live tag field is a pretty telling sign of how rough these AI-generated content pipelines can be.

Wrap-up

• OpenAI banned accounts tied to IUVM, an Iran-based influence operation
• IUVM used AI models to generate and translate articles, headlines, and site tags, spreading pro-Iran, anti-Israel, and anti-US content in English and French
• Content was hosted on iuvmpress.co (successor to a domain seized by the FBI in 2020) and iuvmarchive.org
• OpenAI rated it Category 2 on the Breakout Scale: real activity, but limited actual reach
• If you follow AI safety or disinformation research, this is a concrete case worth knowing about!