OpenAI kills Russia and Iran covert AI operations planting fake news
AI propaganda just graduated from social media bots to polished long-form articles published by real media outlets.
Policy · Source: CNN
What happened
OpenAI just banned two massive covert influence operations from Russia and Iran. These groups used ChatGPT to run sophisticated propaganda campaigns across the globe. They did not just spam social media with cheap bots. They used AI to create fake personas, draft polished content, and trick legitimate news outlets into publishing their lies. The goal was to launder geopolitical messaging through authentic channels.
The Iranian group created seven fake journalists. These AI-generated writers pitched and published roughly 100 long-form articles in outlets like the LA Progressive, Daily Kos, and the UK-based Middle East Monitor. The Middle East Monitor alone has nearly two million Facebook followers. They wrote in Farsi and English, pushing narratives that framed the United States as an aggressor, Israel as an unreliable ally, and Iran as a resilient victim. Once the articles were published, the operatives used ChatGPT to generate fake social media comments to make the stories look viral.
The Russian operation, nicknamed Dark Clark, was even more complex. OpenAI rated it a Category 5 threat on their breakout scale, marking the highest severity disruption they have ever handled. Operating via VPNs to bypass OpenAI restrictions, the Russians ran a fake Latin American think tank called the Social Research Center. They tricked real people into working for them. They manipulated schools in Peru into hosting events for controversial Ukrainian nationalist Stepan Bandera, which even prompted a response from a Polish politician. They also spread fake stories about the Argentine president buying Cartier collars for his dogs. Ironically, the Russian operatives mostly used ChatGPT to write status reports to their bosses, sometimes taking credit for events they had nothing to do with.
Key facts
- 7 — Fake journalist personas created by the Iranian operation
- 100 — Approximate number of articles published by the fake Iranian journalists
- Category 5 — Severity rating OpenAI gave the Russian Dark Clark operation
- 30 — Covert influence operations exposed by OpenAI over the past two and a half years
Why it matters
The baseline for trust on the internet just dropped to absolute zero. If you build content platforms, news aggregators, or social networks, your defense mechanisms are already obsolete. Bad actors are no longer struggling with language barriers or relying on cheap click farms. They are using the exact same language models you use to write perfectly fluent, highly persuasive long-form content. They are passing basic editorial checks at established publications. This means the bottleneck for propaganda is no longer content creation. It is distribution, and AI is helping them hack that too.
We are entering an era of automated narrative laundering. AI allows foreign adversaries to build sophisticated false fronts at a fraction of the traditional cost. When legitimate outlets unknowingly publish AI-generated propaganda, it gives the content a massive veneer of credibility. This forces AI companies like OpenAI to become geopolitical arbiters. They are spending massive resources hunting down state-sponsored actors who use VPNs and commercial APIs to hide their tracks. If you are building foundational models, you are now in the cyber warfare business whether you like it or not.
For builders
Identity verification is now mandatory
Anonymous or pseudonymous publishing is a massive liability. Platforms that allow unverified community posts are prime targets for AI personas. You will pay the price in reputation if you do not verify who is actually writing on your site. Build strict identity checks into your onboarding flow.
AI safety teams are counterintelligence
OpenAI is actively hunting state actors and sharing data with government authorities. If you build wrapper apps or fine-tune models for content generation, expect heavy regulatory scrutiny. You must build robust abuse-detection pipelines or risk being shut down for enabling foreign interference.
The death of the cold email pitch
Iranian operatives successfully pitched human editors using AI-generated emails. Editors and content managers will soon stop accepting unsolicited submissions entirely to avoid being duped. Builders have a massive opportunity to create tools that cryptographically verify human authorship to solve this incoming trust crisis.
My take
I have been saying this for months. We are worried about AI destroying the world, but the real threat is AI destroying our shared reality. When a fake journalist can write a fake article that gets published by a real editor, the internet is fundamentally broken. These state actors are using our own tools against us to launder lies. Build tools for truth, or get out of the way.
Original reporting: CNN. This is my rewrite and opinion.