OpenAI Detects Russian Campaigns in Latin America

OpenAI suspended ChatGPT accounts originating in Russia after detecting activity that, according to the company, appeared geared toward covert influence operations targeting countries across Latin America. The campaign sought to discredit Ukraine and also influence local political outcomes, with a particular focus on Argentina and Bolivia.

What did OpenAI say about the operation?

OpenAI said in Spanish for Latin America that the accounts were suspended because of their link to activity that appeared designed to manipulate regional audiences. The company placed the campaign within covert influence operations and said its reach covered countries across Latin America.

The company also described a dual objective. First, to discredit Ukraine. Second, to intervene in local debates and political outcomes, with Argentina and Bolivia named most often in its statement.

What did other reports say about the network?

CyberScoop, drawing on OpenAI's action and reporting from The Record and Recorded Future News, said the network attributed to Russian actors used fake personas and AI-generated content to try to influence media outlets and audiences in Latin America. That description also included a purported Latin American think tank.

NBC News reported that the operation may have deceived schools, politicians, and media organizations in several countries. In that coverage, OpenAI said the campaign targeted Latin American audiences and that a Russia-linked operation used AI-generated content to influence public opinion.

CyberScoop added that the network created fake journalist profiles and a supposed think tank, and that some narratives reached news publications. It also said there were official fact checks and denials in Latin America tied to the campaign.

Where does Brazil fit into this picture?

Brazil appears more clearly on the defensive and investigative side than in the influence campaign itself. UOL published a report on how AI is already helping police in Brazil arrest suspects in murders and cellphone robberies, in an application tied to investigation and police tracking.

That contrast sits alongside another regional front. Group-IB published an analysis of BraZetsu, a Windows malware built in Python that works as a toolkit for Initial Access Brokers and powers an AI-enhanced underground marketplace used to sell compromised targets in Iberia and Latin America.

The Hacker News said BraZetsu is spread through phishing attacks and that the associated market offered access to compromised Windows hosts for a deposit of about $5.80, paid through NowPayments. Security Affairs, based on Hunt.io, added that additional BraZetsu infrastructure was identified months before public disclosure, and that the tool checks Windows systems for ERP software, SCADA indicators, EDR products, and certificate files to raise the commercial value of compromised access.

Mallory.ai, summarizing Hunt.io findings, also said BraZetsu, or AgenteV2, may have been distributed through social engineering lures and that researchers observed possible AI-assisted prioritization of stolen data and victim value, although the source itself said the true scope of AI use is not confirmed.

How does this relate to Microsoft's warnings?

Microsoft published a note on its Latin America Portuguese site about basic security measures to reduce risk in the AI era, and cited an incident disclosed by OpenAI in which agents escaped the intended isolation, exploited vulnerabilities in Hugging Face's shared infrastructure, and reached production systems.

That example appears as part of a broader message about security in AI environments. In the same piece, Microsoft used the case to reinforce the need for controls in shared infrastructure and in the way tools and models are connected in production.