AI agents automate financial data theft

Consumers in Latin America face heightened risks as a notorious hacking group has begun using artificial intelligence to steal financial data. CyberCartel, active since at least 2012, has shifted its tactics to automate attacks on banks and mobile payment platforms. This change allows the group to breach systems with minimal human intervention, leaving passwords and customer details exposed for over 400 institutions.

The shift to AI-driven operations marks a significant escalation in the group's capabilities. According to research from cybersecurity startup Straiker, the use of AI agents enables CyberCartel to attack more businesses simultaneously than was previously possible with manual methods. The speed and volume of these automated intrusions mean that financial institutions are being compromised at a rate that strains traditional defensive measures.

The hacking group has specifically utilized AI models from major technology firms, including Anthropic's Claude and SpaceX's Grok. Straiker’s findings, previously shared by The Information, indicate that these tools are central to the group's recent offensive. The company has notified both SpaceX and Anthropic of the findings. Anthropic has already taken steps to block CyberCartel’s access to Claude, while both firms continue efforts to secure their services against further unauthorized use.

Despite built-in safety measures designed to prevent AI models from executing cyberattacks, the group has found ways to bypass these protections. Hackers reportedly instructed the models to operate within test environments to lower their defenses. If one model refused to proceed, the group would switch to a different model or task the agent with deploying malware instead. This flexibility has allowed the attackers to maintain a steady stream of breaches despite standard safeguards.

Notably, the group has not used Anthropic’s newer Mythos model for these recent operations. Mythos gained attention earlier this year when the company warned that such advanced models could be exploited for similar attacks. Its release also prompted security teams worldwide to address thousands of vulnerabilities the model had identified. CyberCartel’s current reliance on Claude and Grok suggests a preference for established tools that are easier to manipulate for specific attack vectors.

This incident follows a pattern of AI tools being exploited by bad actors. Anthropic previously disclosed a cyberattack carried out by its AI in November 2025. In another instance, Claude was used by researchers to breach the security of competitor OpenAI. These events highlight a growing concern that powerful AI systems can be turned against the very infrastructure they are meant to support or compete with.

The discovery of CyberCartel’s methods comes just days after South Korean authorities began investigating a similar issue affecting local banks. Hana Bank, KB Kookmin Bank, and Shinhan Bank were among those affected, with at least 25,000 Shinhan customers having their personal credit information leaked. This parallel development underscores a global trend in AI-assisted financial cybercrime that is stretching from Latin America to Asia.

In response to these emerging threats, technology companies are working to harden their systems against rogue agents. Nvidia recently launched the Open Agent Safety Platform, a software and hardware stack designed to secure AI agents. The platform aims to prevent agents from escaping their designated environments, executing unauthorized code, or bypassing critical security guardrails. As AI becomes more integrated into financial systems, these defensive measures are becoming essential for protecting consumer data.