OpenAI said it disrupted a Russian influence operation that used a fabricated think tank leader to plant propaganda in Latin American media, along with an Iranian campaign that ran fake journalists, CyberScoop reported Thursday, citing the company's latest threat report. OpenAI rated both operations well above most of the activity it tracks for how far their content spread.

What Happened

OpenAI named the Russian effort Dark Clark after Mia Clark, a fake AI persona the operators presented as head of the Social Research Center, a supposed Latin American think tank. The persona was used to push material into news coverage. Most of the content aimed to damage Ukraine's reputation in Latin America, but the operation also weighed in on contested political issues in Argentina and Bolivia.

CyberScoop described several examples. In March, the group spread a false TikTok video claiming Ecuador was recruiting its citizens to fight for Ukraine, along with fabricated audio of a Ukrainian consul insulting Ecuadorians. During anti-government protests in Bolivia in May, it released fake audio of a state water company employee claiming the government was about to declare a state of emergency and cut off water to La Paz. The Bolivian government issued an official denial.

RelatedSuspected Chinese Spies Exploited BeyondTrust Remote Support Flaw Since August, Google Says

OpenAI said the evidence pointed to Russian operators directing the persona's work. In reports generated through ChatGPT, operatives discussed pay scales for staff as well as hiring and firing decisions. The report does not tie the activity to a specific Russian government or intelligence agency, though it says the operators showed particular interest in Politology, which CyberScoop described as a successor to the Wagner Group mercenary force. OpenAI called Dark Clark the most complex attempt to run a front identity that it has disrupted in the past two and a half years.

The Iranian operation used at least seven fake journalist identities to publish or syndicate nearly 100 articles across roughly a dozen small and midsize outlets that cover international affairs, focusing on U.S.-Iran relations and Middle East events. The stories often drew few comments or social media reactions. OpenAI did not name who ran it, saying it resembled "a commercial actor running a for-hire influence campaign."

Why It Matters

OpenAI rates the operations it disrupts on an impact scale of 1 to 6, and few rise above 1 or 2. It gave the Russian campaign a 5 and the Iranian campaign a 4. The higher scores reflect content that broke out of the operators' own accounts into mainstream audiences, prompting fact-checks from media organizations and denials from government officials in Latin America.

That distinction matters because many AI-assisted influence efforts OpenAI has reported in the past produced large volumes of content that few people saw. Dark Clark shows a different pattern: a single, carefully maintained persona with a plausible institutional cover, used to feed material into news reporting, combined with fabricated audio timed to moments of public tension.

Assorted national flags mounted on a wall
Illustrative photo: Unsplash. Not a photograph of the reported event.

Context

OpenAI publishes periodic threat reports describing state-linked and commercial actors that misuse its models and the steps it has taken to disrupt them. The reports give a partial view: they cover activity on OpenAI's own services, and the company relies on what operators reveal in their conversations with ChatGPT, such as the internal staffing discussions that pointed to Russian control of Dark Clark.

What to Watch

CyberScoop's account does not identify which mainstream outlets carried Dark Clark material or which outlets published the Iranian articles. Whether those publishers review or retract the content, and whether social platforms such as TikTok remove accounts tied to the Mia Clark persona and the fake journalists, will show how much of the operations' reach survives OpenAI's disruption.