OpenAI exposes Russian disinformation network directed against Ukraine
OpenAI has discovered and blocked a network of accounts from Russia that used ChatGPT to conduct covert information operations in Latin American countries. One of the main goals of the campaign was to discredit Ukraine.
To spread disinformation, its organizers recruited real journalists, created fake documents, and managed a research organization whose employees likely did not know about the Russian origin of the project.
This is stated in an OpenAI investigation published on October 8. The company named the discovered operation Dark Clark.
According to researchers, the network operated mainly in Latin America. Its participants tried to worsen Ukraine's international reputation, discredit military service in the Ukrainian army, and influence domestic political processes in certain countries, including Argentina and Bolivia.
Most operators communicated with ChatGPT in Russian. To bypass access restrictions to the service from the territory of the Russian Federation, they used VPN.
A distinctive feature of the operation was the use of a real analytical organization as a cover. According to OpenAI's conclusions, the Russian organizers controlled the Social Research Center (SRC) - a research platform that operated in Latin America.
Internal operator reports mentioned personnel decisions, salary levels, hiring of employees, and research projects. Management of the organization was carried out, in particular, through a fictitious persona named Mia Clark.
At the same time, SRC employees who worked directly in the region most likely did not suspect the organization's connection to the Russian information operation. They conducted interviews with experts, prepared research, and published analytical materials.
OpenAI specialists found more than 60 articles on the SRC website, most of which were original. This indicates an attempt to create a long-term influence network, not just a short-term social media campaign.
One of the most telling episodes involved an attempt to provoke a conflict between Ukraine and Poland.
According to internal operator reports, in May 2026 they created a fraudulent email address in the name of the education authority of the Peruvian capital, Lima. From this address, schools were sent instructions to organize events dedicated to Ukraine with mentions of Stepan Bandera.
Subsequently, the campaign organizers claimed to have used these events to spread publications about Ukraine allegedly promoting ultranationalist ideology.
OpenAI investigators found materials matching this scenario in Peruvian and Polish media, as well as in an English-language outlet in Hungary. Some publications were later removed.
Some articles included a reaction from a Polish member of the European Parliament who spoke in favor of restricting entry to Poland for individuals accused of anti-Polish activities. Thus, according to OpenAI's assessment, the Russian information operation was able to fuel real tensions between Ukraine and Poland.
In another case, operators created a fake audio recording attributed to a Ukrainian consul in Ecuador. It was used to spread fictitious offensive statements against the local population.
The network also spread false reports about the alleged recruitment of Ecuadorian citizens to participate in the war in Ukraine. To increase persuasiveness, they used forged documents and video materials with the branding of real news media.
OpenAI emphasized that the use of artificial intelligence was not limited to creating fakes. Operators devoted a significant portion of requests to preparing internal reports about their work results. In some cases, they even attributed information events to themselves that they had nothing to do with, likely to exaggerate their effectiveness in front of leadership.
Meanwhile, researchers were able to independently confirm some episodes. Some fabricated messages spread so widely that they were debunked by fact-checkers and government officials.
On the six-level scale of information operation reach, OpenAI assigned Dark Clark the fifth category. The company noted that this was the first campaign of this level it had exposed since the publication of relevant reports.
Researchers also found possible connections between certain episodes and Russian structures Politology and La Compania, which had previously been linked to the influence network formed around Wagner PMC founder Yevgeny Prigozhin. At the same time, OpenAI does not claim that all results reported by the operators have been confirmed.
Following the investigation, the company blocked ChatGPT accounts associated with the campaign and passed information to competent authorities. Separately, OpenAI reported the termination of an Iranian information operation whose participants published material under fictitious journalist names in foreign online publications.
Source: OpenAI
