The fake byline is still a security problem. AI makes it easier to scale.

Borrowed credibility

OpenAI says it recently banned two influence operations originating in Russia and Iran that used ChatGPT to support deceptive front identities. In a report published on 8 October, it describes fake journalist personas pitching articles and a purported research organisation targeting audiences in Latin America. [1]

The company says the Iran-origin operation used seven journalist identities and that it found almost 100 articles published or syndicated under their bylines. Its account of the Russia-origin operation describes content creation and internal reporting, with some operators taking credit for activity they did not carry out. [1]

Our reading is that the trust problem begins when an invented contributor borrows the credibility of an established outlet. Fluent language may make that hand-off easier, but it does not establish who wrote an article, why it was submitted, or whether its claims are true.

How to read an impact score

To assess reach, OpenAI uses the Breakout Scale, a framework published by Brookings in 2020. It separates six categories according to how an operation moves between communities, platforms, mainstream media and influential public figures. [2]

The framework’s purpose is to compare observable distribution and potential impact. Its paper warns that a message being repeated by a policymaker does not necessarily mean it succeeds: the message could be rejected. Exposure and persuasion are different things. [2]

OpenAI assesses the Russia-origin case at Category 5 and the Iran-origin case at Category 4. Those are the company’s applications of the framework, rather than independent scores produced by Brookings for these cases. [1] [2]

What the story changes for publishers

Our reading is that the vulnerable point is the hand-off from an unknown contributor to a trusted publication. Fluent copy, a plausible biography and a convincing pitch can all look like ordinary editorial work. A newsroom has to establish who is making a claim and what evidence supports it.

For this publication, that means keeping source links attached to factual claims, recording where images come from, and making the editorial responsibility visible. These are process choices we can test; they are not evidence that South African outlets were involved in the disclosed operations.

Our conclusion is about accountable review. When content arrives behind a borrowed identity, fluency is a weak basis for trust. A larger pile of polished articles should never be confused with a stronger body of evidence.

AI-assisted research and writing, with a separate source review. Case findings are attributed to OpenAI. The Brookings paper supplies methodology, not independent investigation of these cases.