OpenAI bans Russian 'Dark Clark' and Iranian 'Bogus Bylines' ops

OpenAI banned accounts behind two foreign influence operations, including an Iranian network that used seven fake journalists to plant nearly 100 fabricated articles in real online outlets worldwide.

The combined takedown ranks among the larger disclosures OpenAI has issued. The pair of operations, code-named "Bogus Bylines" and "Dark Clark," show how state-aligned actors now treat AI chatbots as cheap, scalable writing staffs.

What did the Iranian "Bogus Bylines" operation actually do?

The Iranian campaign, code-named "Bogus Bylines" by OpenAI, ran seven fake journalists as personas. Those personas placed close to 100 articles through online outlets worldwide, the company said in its latest threat report. The accounts wrote and distributed content at scale.

OpenAI didn't name the outlets or the personas in its disclosure. The network published across multiple languages, the company added. The fake journalists used AI tools to draft and translate their pieces.

Seven personas let the network look like a working newsroom. Each persona posted on a steady schedule, OpenAI noted. The accounts rotated topics so no single beat looked overused.

Bogus Bylines shows how cheap and effective a small team has become. The whole operation fits inside one threat category at OpenAI. It does, however, touch real newsrooms and real readers across several regions.

What made the Russian "Dark Clark" campaign different?

The Russian campaign, code-named "Dark Clark," spread disinformation across Latin America. It drew reactions from politicians in the region, OpenAI said. The company rated Dark Clark at category 5 on its six-point threat scale. That is the first category 5 rating in OpenAI's reporting history.

Dark Clark used AI tools to generate content at scale, OpenAI's report added. The campaign posted across social platforms and messaging apps. Analysts at OpenAI said the operation showed clear signs of state backing.

The category 5 rating matters more than the campaign name. OpenAI uses a six-point scale for influence operations, with 6 the highest. Categories 1 through 4 cover low-impact spam and single-platform abuse. Category 5 covers operations that move real people or distort real events. Category 6 hasn't been publicly disclosed by OpenAI to date.

Dark Clark earned the higher rating because it pulled responses from sitting politicians, OpenAI noted. That kind of feedback loop is rare in state-linked influence work. It makes false stories harder to dismiss as fringe chatter.

A piece that prompts a real elected official to issue a statement creates a second wave of coverage. Other outlets pick up the response. Readers see the false story repeated in context. OpenAI's report describes that pattern across several Latin American markets.

How does OpenAI's threat scale work?

OpenAI's six-point threat scale runs from nuisance spam at the low end to coordinated state operations at the high end. Most disclosed campaigns land in the middle of the range. The category 5 rating puts Dark Clark one notch below the worst case OpenAI has ever named.

The scale grades more than influence work. It covers account abuse, content farms, and coordinated inauthentic behavior across OpenAI's products. A category 5 score is rare, OpenAI noted. It signals an operation that moved real people, not just real pixels.

Dark Clark's reach into Latin America is also notable. Most public influence disclosures focus on U.S. and European targets. A category 5 operation aimed at Latin American politics stands out. OpenAI didn't name which countries or which politicians reacted.

The Iranian campaign didn't draw a category 5 rating, OpenAI added. The two cases sit at different points on the scale. Bogus Bylines spread wider in volume. Dark Clark landed harder in real-world impact.

Why do these disclosures matter now?

Two operations in one disclosure is unusual. OpenAI has named multiple campaigns before, but rarely at this pace. The back-to-back takedowns suggest the company is watching state-aligned use of its tools more closely than it did a year ago.

The news also lands as newsrooms struggle with AI-generated submissions. A fake byline backed by AI tools can pass basic editorial checks. Editors who vet contributors by writing samples now face a harder task.

For readers, the lesson is simple. A byline tells you less than it used to. A real-looking article from an unknown writer deserves a second look, no matter where it runs. The platform no longer guarantees the writer.

For policymakers, the pair of cases shows how AI tools expand the reach of small influence teams. Seven fake journalists can now produce content that used to require a full newsroom. A handful of operators can move real politicians.

The disclosures also raise a harder question. What happens when a category 6 case lands? OpenAI hasn't named one yet. Dark Clark sits one notch below. The next disclosure, though, may not stay at 5.

OpenAI's trust and safety team has stepped up its reporting cadence since 2024. The company now publishes threat reports more often than once a quarter. That rhythm helps outside researchers and platforms track patterns across operations.

OpenAI said it will keep publishing threat reports on a rolling basis. The company also said it will share account data with outside platforms. Neither campaign reached category 6. The next disclosure, though, may not stay at 5.