400+ Latin American Institutions Breached by AI-Driven Attacks

A major cybercrime group has compromised hundreds of financial entities using artificial intelligence.

The hacking collective known as CyberCartel has targeted banks and mobile payment applications across Latin America. According to research from the cybersecurity startup Straiker, the group has stolen passwords and customer details from more than 400 institutions. Straiker shared its findings with The Information. The attacks have resulted in unauthorized access to personal bank accounts with very little direct input from human operators.

CyberCartel has been a threat to Latin American businesses since at least 2012. In recent months, however, its tactics have changed. The group now employs AI agents to automate its attacks. Straiker reports that this shift allows the group to target a volume of businesses at speeds human hackers could not match.

The tools used by CyberCartel include Anthropic's Claude and SpaceX's Grok. The group has also used an agent from the Chinese lab Moonshot AI. Straiker is attempting to contact that company regarding the matter. Both SpaceX and Anthropic received Straiker's research and are working to cut CyberCartel off. Anthropic has already blocked access to Claude for the group.

Security measures on AI models are intended to stop such misuse. CyberCartel bypassed these protections by claiming it was operating in a test environment. If a model refused to cooperate, the group simply switched to another. If that model also refused, the hackers directed the agents to deploy malware instead.

Anthropic's Mythos model does not appear to be part of these recent attacks. Earlier this year, Anthropic warned that new models could be used for attacks similar to CyberCartel's. The release of Mythos also led security teams worldwide to address thousands of vulnerabilities the model identified.

This is not the first time Anthropic's AI has been used for malicious purposes. In November 2025, the company disclosed that its AI was used in a cyberattack. Later, Claude was used to break the security of OpenAI. That specific hack was conducted by researchers rather than criminals.

Straiker's research follows news from South Korean authorities regarding a similar issue. Officials there are investigating attacks affecting local banks. Hana Bank, KB Kookmin Bank, and Shinhan Bank were affected. At least 25,000 Shinhan Bank customers had their personal credit information leaked.

AI companies are aware of the risks their products pose. Nvidia launched the Open Agent Safety Platform to address these concerns. The software platform and reference system design aim to secure rogue agents. The goal is to prevent agents from escaping sandboxes, executing unauthorized code, or accessing critical infrastructure.