AI Agents Accelerate Latin America Bank Attacks

A notorious hacking group is using artificial intelligence to speed up attacks on financial institutions across Latin America.

According to research from cybersecurity startup Straiker, the group known as CyberCartel has deployed AI agents to automate its operations. The targets have included banks and mobile payment applications, resulting in the theft of passwords and customer information. This activity has also enabled access to individual bank accounts with minimal manual effort from the criminals. The findings from Straiker were shared by The Information.

While CyberCartel has been a persistent threat targeting Latin American businesses since at least 2012, its recent adoption of AI has significantly enhanced its capabilities. The group can now compromise organizations at a pace that was previously impossible when relying on human operators alone. Straiker has shared its research with both Anthropic and SpaceX, and both companies are taking steps to restrict CyberCartel's access to their systems. Anthropic has already successfully blocked access to Claude. Straiker is currently attempting to contact Moonshot AI regarding the hacking group's use of its agent.

Despite standard safety protections designed to prevent misuse, these safeguards have proven relatively simple to circumvent. CyberCartel reportedly bypassed restrictions by claiming the models were operating within a test environment. If one model declined to proceed, the group would switch to a different model. Should that also refuse, the AI agents were tasked with deploying malware instead. Notably, CyberCartel does not appear to be using Anthropic's Mythos model for these recent attacks, despite the company's earlier warnings that models like it could be utilized for such activities. Anthropic has previously disclosed instances of its AI being used in cyberattacks, including a November 2025 incident and a later breach of OpenAI's security conducted by researchers.

This research follows closely on the heels of similar incidents reported elsewhere. Authorities in South Korea are currently investigating attacks affecting local banks, including Hana Bank, KB Kookmin Bank, and Shinhan Bank. At least 25,000 customers at Shinhan Bank have had personal credit information exposed. Regarding the broader security of AI systems, Nvidia has launched the Open Agent Safety Platform. This hardware and software stack is designed to secure rogue agents and prevent them from escaping sandboxes, executing unauthorized code, or gaining access to critical infrastructure.

As AI companies become increasingly aware of the potential for their products to be exploited by bad actors, efforts to secure these systems are ramping up. The development of tools intended to prevent unauthorized access or bypass guardrails is becoming a priority for providers like Anthropic and Nvidia. Further updates on how these companies plan to address these vulnerabilities are expected as the situation develops.