OpenAI bans Russian and Iranian influence networks: 7 fake journalists and ~100 articles

OpenAI has exposed and shut down two influence operations, one Russian and one Iranian, and blocked all accounts linked to them. The Russian campaign “Dark Clark” spread disinformation across Latin America and prompted reactions from politicians. As a result, it received category 5 on OpenAI's six-point scale, the highest rating in the history of the company's reports. The Iranian “Bogus Bylines” was subtler: seven invented journalists placed nearly 100 articles in online publications in various countries.

Both cases show how the practice of information operations has shifted. Operators used to focus on bots and mass posting on social media. Now they try to get onto the pages of real media outlets, where a text carries a very different level of trust.

Dark Clark: disinformation that politicians noticed

According to OpenAI, the Russian campaign targeted a Latin American audience. The record score reflects impact, not volume. The scale measures how far an operation travels beyond its own bubble audience and whether it triggers a real-world response. Politicians' reactions are a strong signal in that sense: the fake stopped circulating among bots and entered the public agenda.

In the several years that OpenAI has published such reports, the typical operation stayed at the low levels: texts were generated and published, but almost nobody read them. “Dark Clark” broke that pattern. For the industry, this is a reason to revisit the assumption that AI propaganda mostly remains noise.

Bogus Bylines: seven authors who did not exist

The Iranian operation works differently. Instead of a loud social media campaign, its participants created seven fictitious journalists and used them to push material into online publications. Nearly 100 publications worldwide average out to 14 texts per nonexistent author. To an editor, such a freelancer looks like an ordinary person with a portfolio, and the scheme depends on exactly that.

Key figures from the report:

• Dark Clark (Russia): target region is Latin America, category 5 out of 6, the first such rating in OpenAI's reports.
• Bogus Bylines (Iran): 7 fake journalists, nearly 100 articles, publications reached worldwide.
• Measures: blocking of all identified accounts linked to the operations.
• Injection channel: real news outlets, not just social media.

What we know and what we don't yet

The public summary of the report is terse. Several questions remain open: which publications ran the material, what share of the texts a model wrote, and how many of them editors have already removed. It is also unclear how much of the chain OpenAI can see. The company observes only what happens on its own platform, and the texts could have been refined by hand or in other services. It is worth remembering, then, that blocking accounts takes away one tool but does not end the operation itself.

How newsrooms and readers can verify authors

The main risk lies on the receiving end. If AI makes plausible text cheap, the filter has to become verification of the author's identity, not the style of the writing. In practice, that means several things:

• Check new freelancers against independent sources: publications in other outlets, profiles, the possibility of live contact by video.
• Run the author's photo through reverse image search, since generated portraits often have no history.
• Look at the coherence of the portfolio: topics that line up sharply with one state's interests should raise questions.
• For readers: be cautious about little-known authors whose texts suddenly spread across many small sites at the same time.

The OpenAI report confirms that model developers' platforms are gradually becoming one source of data on such operations. But responsibility for publication stays with the newsroom, and that is where this filter should kick in first.