AI helped influence campaigns reach real news outlets
October 9, 2026
OpenAI banned Russian and Iranian networks that used ChatGPT for false identities and media pitches. For the first time, the company rated one operation at reach category 5.
What this is about
OpenAI disclosed two covert influence operations from Russia and Iran on October 8, 2026, and banned the associated ChatGPT accounts. The important part is not simply that AI can produce propaganda text, but how conventional deception and AI worked together: a Russian group allegedly used unwitting people for a supposed think tank, while an Iranian group operated seven fictional journalist personas and pitched their work to real editors.
OpenAI rated the Russian operation as category 5 on the six-level Breakout Scale. This is the highest rating the company has assigned to an operation it disrupted since its reporting began in early 2024. The Iranian operation reached category 4.
What the influence operations actually did
The Russian operation, which OpenAI calls “Dark Clark,” used ChatGPT mainly for internal reports, translations and campaign material concerning Latin America. According to the company, it controlled a “Social Research Center” through a false identity. Local staff apparently did not know that a Russian group was behind it. The operators also claimed to have spread forged emails and stories; some matching reports, fact checks and official denials could be found publicly.
The Iranian operation used seven supposed journalists. It submitted long-form articles to small and medium-sized online outlets and generated comments about topics surrounding the US-Iran war. OpenAI explicitly says that not all published material came from its models.
Why it matters
The campaigns did not rely only on anonymous social-media accounts. They tried to borrow the credibility of existing publications. An article on a real news site appears more trustworthy to readers and search engines than a post from a new, unknown profile.
OpenAI has disclosed 30 covert influence operations since early 2024. Its analysis says campaigns that place material in external publications tend to achieve greater reach than networks relying only on their own sites or social platforms. The case exposes a practical weakness: newsrooms may scrutinize copy more closely than the real identity, funding and client behind a freelance contributor.
In plain language
The method resembles someone handing a parcel to a trusted neighbor instead of sending it under their own name. The contents stay the same, but the recipient is more likely to open it because they trust the sender. AI helped with fast writing and translation; the borrowed identity was the core deception.
A practical example
A small newsroom receives a polished guest article from a supposed correspondent, complete with source links. Two editors check the figures and language, see no obvious errors and publish it. Only later do they learn that the profile photo, biography and sponsor were fabricated. AI can help a group produce ten localized versions in parallel, but that alone does not prove that every version was published or influenced anyone.
Scope and limits
- Evidence about accounts and internal prompts comes mainly from OpenAI; independent investigators have not verified every detail.
- A Breakout rating reflects potential reach and distribution methods, not proven changes in public opinion or election outcomes.
- AI was an accelerator, not the sole cause. False identities, recruited people, email tactics and editorial failures mattered too.
SEO & GEO keywords
OpenAI, ChatGPT, influence operation, disinformation, Dark Clark, Iran, Russia, fake journalists, media manipulation, Breakout Scale, Latin America
💡 In plain English
Russian and Iranian groups used ChatGPT to make covert campaigns look more professional. The key risk was that fabricated identities placed content in real media outlets.
Key Takeaways
- →OpenAI banned accounts tied to influence operations from Russia and Iran.
- →The Russian operation became the first in OpenAI's reporting to receive a category 5 reach rating.
- →The Iranian operation used seven fabricated journalist personas.
- →Placements in real publications mattered more than social-media volume alone.
- →OpenAI does not claim proven effects on opinions or elections.
FAQ
What is Dark Clark?
Dark Clark is OpenAI's name for a Russian influence operation active in Latin America that apparently controlled a supposed think tank.
Did AI publish the articles by itself?
No. AI supported writing, translation and internal reporting; people, false identities and real newsrooms formed the distribution path.
What does category 5 mean?
On the six-level Breakout Scale, category 5 indicates high potential reach through established media channels. It does not prove a specific political effect.