CyberCartel uses AI agents to automate bank theft in Latin America

A hacking group called CyberCartel has begun using artificial intelligence agents to automate attacks on banks and mobile payment apps in Latin America, according to research from the cybersecurity startup Straiker. The findings were recently shared by The Information. The group has operated in the region since at least 2012, but the new tools allow it to target more organizations at a pace that exceeds what human-driven hacking can achieve.

Straiker identified specific models used in the operation, including Grok from SpaceX, Claude from Anthropic, and an agent from Moonshot AI. The research indicates that passwords and customer data have been stolen, and individual bank accounts have been accessed with minimal direct involvement from the criminals. The group reportedly told the AI models they were operating in test environments to bypass standard safety refusals. When one model declined to continue, the hackers switched to another or assigned the agents to deploy malware instead.

Anthropic has already blocked CyberCartel's access to Claude. Straiker passed its findings to SpaceX and Anthropic, and both companies are working to cut the group off from their services. Straiker is still trying to reach Moonshot AI. The incident demonstrates how fragile the safeguards built into AI models can be when attackers know how to phrase their requests.

The Straiker report notes that the group has not used Anthropic's Mythos model. That model drew attention earlier this year when Anthropic warned that hackers could exploit it for similar attacks. The release of Mythos also prompted security teams worldwide to rush to fix thousands of vulnerabilities the model had identified. Anthropic is not new to such findings, having disclosed in November 2025 that its AI was used in a cyberattack. Claude was later used to break the security of OpenAI, a rival developer, though that specific operation was carried out by researchers rather than criminals.

The report arrives days after South Korean authorities began investigating a comparable problem affecting local banks. Hana Bank, KB Kookmin Bank, and Shinhan Bank are among those affected. The personal credit information of at least 25,000 Shinhan Bank customers has been leaked in that incident. Industry players are responding to these risks. Nvidia recently launched the Open Agent Safety Platform, a software and hardware design intended to contain rogue AI agents. The platform aims to stop agents from escaping sandboxes, executing unauthorized code, or bypassing guardrails.