Skip to content
by InfinyAI ES

IA en un minutoNewsSecurity

Security

OpenAI exposes a Russian operation that used ChatGPT and reached schools in Peru and Ecuador

IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds

OpenAI has shut down ChatGPT accounts run from Russia by a covert campaign in Latin America, largely aimed at Ukraine, which it calls Dark Clark. The operators barely used it to produce propaganda and relied on it instead for internal paperwork, while running a fake think tank whose local staff, according to OpenAI, did not know who they worked for. With fake emails they got, by their own account, schools in Peru and Ecuador to hold events.

OpenAI has banned a group of ChatGPT accounts that originated in Russia and belonged to a covert influence operation aimed at Latin America. The company calls the operation "Dark Clark" and ranks it in Category 5 of the Breakout Scale, which scores such operations from 1 to 6. It is the first Category 5 operation OpenAI has disrupted since it started publishing these reports.

OpenAI says a large share of the activity was meant to hurt Ukraine's image in the region, while another share sought to sway local politics, above all in Bolivia and Argentina. Russian was the language most operators wrote in; one used Spanish but also seemed to be in Russia. Because OpenAI does not permit access from Russia, they connected to ChatGPT through VPNs. Their main use for it was drafting internal reports about their campaigns for a superior whose identity is unknown, and in most cases OpenAI did not see them use it to create campaign content.

Those reports also covered a supposed research center in Latin America, the Social Research Center, which the operators ran under a false identity, "Mia Clark", the source of the operation's name. The available evidence, OpenAI says, suggests the staff it employed in the region had no idea their employer was a Russian group. The center's website describes its subject as Latin America's Indian diaspora. OpenAI counted more than 60 articles there, the vast majority original.

In their reports the operators took credit for hoaxes that circulated in the region. Some had earlier been pinned on "Politology" or "La Compania", the names used for a Russian entity described as a successor to the Wagner Group.

The most striking episode took place in Peru. By their own account, in May this year the operators set up a fake email address imitating Lima's Regional Directorate of Education and asked schools to devote events to Ukraine on May 21, the Day of Cultural and Linguistic Diversity, with references to Stepan Bandera. Some Ukrainians regard this twentieth-century nationalist as a figure of independence, whereas Russia and Poland associate him with fascism, collaboration during the war and atrocities. The operators reported that some schools wrote back to say the events had taken place, and even attached photos. Later, stories consistent with their account ran in the Peruvian and Polish press (some have since been deleted), one of them carrying the reaction of a European Parliament member from Poland. According to OpenAI, the hoax drew on historical tensions between Poles and Ukrainians and stoked them at the same time.

In June, again by the operators' account, a second fake email asked schools in Ecuador to stage a ceremony of loyalty to Erik Prince, who formerly led the private military company Blackwater, and to President Daniel Noboa. OpenAI found press reports closely matching that description, along with a detailed denial from Ecuador's education minister.

OpenAI cautions that the impact needs to be judged with care. The operators often claimed credit for things unrelated to them, so their reports cannot be taken at face value, although open sources did corroborate some of what they said.

The same report covers a second operation, run from Iran, which OpenAI has named "Bogus Bylines". Through seven fake journalists, it offered articles on the conflict between the United States and Iran to small and mid-sized digital outlets worldwide, and OpenAI located almost 100 published under those bylines. That operation sits in Category 4. Both succeeded in placing content, not all of it generated with OpenAI's models, in mainstream media outlets. OpenAI has banned the two groups and passed information on both cases to the relevant authorities.

Why it matters · analysis and opinion

What stands out in this case is how little it relied on artificial intelligence. The fake that went furthest was an email impersonating an education authority, a technique as old as email itself, and the chatbot mostly served the operators' internal paperwork. That puts defense back where it has always been: any school, company or newsroom that receives an unexpected official instruction should confirm it through another channel before acting, and small outlets that accept articles from outside contributors have reason to check who is behind the byline. The report also leaves an open question. OpenAI only sees what happens inside its own service, and the operators themselves inflated their achievements to their boss, so the real reach of these campaigns remains hard to measure from the outside.

Official source: OpenAI · Written with the help of AI: how we make the news

AI in your inbox, every day or every Friday

The stories that matter, each one in a minute. With the source for every one.

Choose one or both:

Sign up and you are in: the daily arrives every night and the weekly on Friday mornings. You can unsubscribe from any email.