AI influence operations moved beyond fake accounts and into real headlines

OpenAI says it disrupted two covert influence operations originating in Iran and Russia. The Iranian network created seven fake journalist personas and pitched long articles to small and medium online outlets. The Russian operation allegedly used unwitting people in Latin America to run an apparent think tank. The primary page returned a 403 error during direct verification, so this account relies cautiously on its indexed passages and related reporting.

Seven fake journalists and a think tank reached real media outlets

According to OpenAI, the Iranian operation used ChatGPT to edit and pitch nearly 100 articles that appeared under false bylines in genuine publications. The Russian network built a credible-looking institution around people who may not have known the project's real background.

On the IO Breakout Scale from 1 to 6, OpenAI rated the Russian operation Category 5 and the Iranian operation Category 4. The company says the Russian case is the first Category 5 operation it has disrupted since it began reporting.

Operators rent newsroom trust instead of building bot audiences

OpenAI has described 30 covert influence operations over the past two and a half years. Its own observations suggest that campaigns with the greatest potential reach do not rely only on fake social distribution. They try to place material in outlets that already possess readers, editors and reputations.

That changes the control point for newsrooms. Detecting generated text is insufficient because authorship of the sentences is only part of the problem. The origin of the contributor, institution and story entering the newsroom matters more.

The supplier of the model is also the source of the incident report

The findings come from OpenAI, which operated the service used by the actors and decided which accounts to ban. Outsiders therefore cannot fully inspect the underlying data, attribution or the share of content produced by its models. The company itself says not all of the material was generated by its systems.

Categories 4 and 5 measure breakout and potential reach, not a demonstrated change in public opinion. A published article is more serious than a forgotten bot post, but publication alone does not prove impact on an audience.

Contributor checks will determine whether the same tactic works again

The next signal is whether outlets tighten checks on outside contributors, think tank funding and the provenance of pitched articles. Platforms can close accounts after an operation is discovered, while an editor can stop a story before publication.

Independent analysis of the published trail and transparent corrections from outlets that accepted the material will matter too. Without them, the public is left relying mainly on the report from the company whose system the operators used.

Lilith's verdict

The fake journalist no longer shouts from an empty account. They sit in an editor's inbox with a finished article, waiting for a real newsroom to lend them its logo.