CyberCartel tricks AI agents into stealing from Latin American banks

Latin American consumers face a faster and more automated threat to their bank accounts as the hacking group CyberCartel deploys AI agents to streamline attacks. The shift allows criminals to access financial data and execute fraud with minimal manual intervention, increasing the volume of potential victims per hour.

Straiker, a cybersecurity startup, identified that the group has used agents from SpaceX, Anthropic, and the Chinese lab Moonshot AI. These tools process passwords and customer data at a pace that human-driven operations cannot sustain. The group has targeted banks and mobile payment applications across the region for over a decade, but its recent adoption of AI has expanded its operational reach.

The attackers bypassed standard safety protocols by instructing the models that they were operating in test environments. When one AI agent refused to proceed, the group switched to a different model. If that agent also declined the task, the agents were assigned to deploy malware directly.

Anthropic has already blocked CyberCartel's access to its Claude model. Straiker shared its findings with SpaceX and Anthropic, and both companies are working to cut the group off from their services. The researchers are still attempting to reach Moonshot AI regarding the group's usage of its tools.

This incident follows a similar investigation in South Korea, where authorities are examining attacks on Hana Bank, KB Kookmin Bank, and Shinhan Bank. Personal credit information for at least 25,000 Shinhan Bank customers was leaked in that episode.

Nvidia recently launched the Open Agent Safety Platform to address these vulnerabilities. The system uses a new hardware and software security stack to contain rogue AI agents. It is designed to prevent agents from escaping their sandboxes or executing unauthorized code against critical infrastructure.