OpenAI blocked Russian accounts over Bolivia

Update November 4, 2026: OpenAI added details about the operation and said the operators claimed to have spread a fake audio clip in Bolivia in late May, attributed to an EPSAS worker. It also said ChatGPT was used to draft internal reports, create campaign content and prepare reports on the performance of the fictitious Social Research Center platform.

OpenAI said it blocked a set of ChatGPT accounts originating in Russia that were used in a covert influence operation targeting Latin America. The company said the activity showed a particular interest in the political outcomes in Bolivia and Argentina, and that its models were used to draft content and internal reports for the campaign.

What did OpenAI attribute to the operation?

OpenAI said the operation used its models to produce text, documents and internal support material aimed at intervening in Latin American political debates. In the company’s account, the activity was part of a covert influence scheme tied to accounts that OpenAI blocked.

The Record said the Russian operation, which OpenAI called "Dark Clark," appeared to have used ChatGPT to create allegedly leaked documents and audio scripts. That coverage also said some of the material may have circulated online, although it did not provide independent forensic verification of the audio.

What happened in Bolivia?

In Bolivia, the available material points to an effort to influence domestic political debates through audio clips and documents attributed to local institutions, but it does not independently confirm the authenticity of those materials. CyberScoop reported that an audio clip attributed to an employee of the state water company, during the Bolivian protests, triggered an official denial.

OpenAI later added that the operators claimed to have spread a fake audio clip in Bolivia at the end of May, attributed to an EPSAS worker, with a false warning about a water outage in La Paz and a declaration of a state of emergency. The company included that episode within the same covert influence operation.

That coverage also said OpenAI claimed the operations succeeded in inserting narratives into mainstream media. Streamline Feed, meanwhile, reported that Bolivia Verifica had debunked a viral audio clip about an alleged water outage in La Paz, and that the debunking cited the water utility as saying service was normal.

What questions remain about the audio clips and leaked materials?

The main question is the technical attribution of those audio clips and documents. The Record and CyberScoop described materials linked to the operation, but both reports made clear that, based on what was available, there was no independent forensic authentication of the Bolivian audio or official confirmation that it was AI-generated.

Another The Record report said Latin American people associated with the fictitious Social Research Center may not have known they were working for Russian operators. OpenAI added that the operation used ChatGPT to draft internal reports, create content for its operations and prepare reports on the performance of the self-described Social Research Center (SRC) platform. That detail, which helps define the scale of the structure, comes from OpenAI’s description and not from independent verification.