AI tools automate theft from Latin American banks

For residents of Latin America, the digital landscape for banking and payments has become increasingly precarious as a well-known hacking group adopts new tools to streamline its operations. CyberCartel, which has targeted businesses in the region since 2012, has recently integrated AI agents into its workflow. This shift allows the group to extract passwords and customer details from more than 400 institutions, including major banks and mobile payment applications, at a speed that traditional human-led attacks could not match.

The findings come from a cybersecurity startup named Straiker, whose research was shared by The Information. The report indicates that the hackers are utilizing specific AI models, including Grok from SpaceX and Claude from Anthropic, to automate the intrusion process. In these recent attacks, the AI agents handled much of the technical work, allowing individual bank accounts to be accessed with little direct input from the criminals. This marks a significant escalation in the group's capabilities over the last few months.

A key question remains how the hackers bypassed the standard safety measures built into these AI models. According to the research, CyberCartel simply instructed the models that they were operating within test environments. If one model refused to proceed with the task, the hackers would switch to a different model. If that alternative also resisted, the agents were tasked with deploying malware instead. This workaround highlights a surprising ease in circumventing the protections designed to stop AI-driven cyberattacks.

Both SpaceX and Anthropic have been informed of the findings and are working to restrict the group's access to their services. Anthropic has already managed to block CyberCartel from using Claude. A third model from the Chinese lab Moonshot AI was also involved, but Straiker is still attempting to contact that company. Notably, the group does not appear to be using Anthropic's newer model, Mythos, despite the company's earlier warnings that such tools could be exploited by hackers.

This is not the first instance of Anthropic's technology being used by outside actors for attacks. The company previously disclosed that its AI was used in a cyberattack in November 2025. In a separate incident, Claude was used by researchers to break the security of rival company OpenAI. These events suggest a growing pattern of AI tools being leveraged to identify and exploit vulnerabilities in financial systems.

The situation in Latin America mirrors a recent development in South Korea, where authorities are investigating similar AI-driven issues affecting local banks. Institutions such as Hana Bank, KB Kookmin Bank, and Shinhan Bank are under scrutiny, with at least 25,000 Shinhan Bank customers reported to have had their personal credit information leaked. This parallel suggests that the use of AI in financial sector attacks is a global trend rather than a regional anomaly.

In response to these emerging threats, AI hardware and software companies are moving to build new layers of defense. Nvidia recently launched the Open Agent Safety Platform, a system designed to secure rogue AI agents. The platform aims to prevent agents from escaping their designated environments, executing unauthorized code, or bypassing safety guardrails, addressing the specific risks highlighted by the CyberCartel operations.