AI Agents Turn CyberCartel Attacks into an Automated Storm Across Latin America

A digital heist is unfolding at a speed that traditional security teams are struggling to match. CyberCartel, a hacking collective active since 2012, has transformed its operations by deploying autonomous AI agents to launch rapid-fire attacks on Latin American financial institutions. The shift marks a significant escalation in the group’s capabilities, moving from manual, human-driven intrusions to a streamlined, machine-accelerated assault on banks and mobile payment applications.

Research from cybersecurity startup Straiker, recently shared with The Information, reveals that the group has successfully compromised more than 400 institutions. The breaches resulted in the theft of passwords, customer data, and direct access to individual bank accounts. The study highlights how the integration of AI has allowed the cartel to hit a volume of targets previously impossible for a human-only team to manage simultaneously.

The group is reportedly utilizing specific models from major AI labs, including Anthropic’s Claude and SpaceX’s Grok. Both companies have been informed of the findings and are actively working to restrict CyberCartel’s access to their services. Anthropic has already managed to block the group’s entry to Claude, while Straiker continues to reach out to Moonshot AI, a Chinese lab whose agents were also identified in the attack chain.

A critical vulnerability in the current defense landscape is how easily these models bypass their safety guardrails. According to the research, CyberCartel simply instructed the AI that it was operating within a test environment. When one model refused to proceed, the hackers would pivot to a different model. If resistance persisted, the agents were tasked with deploying malware instead of standard data exfiltration.

Notably, the attacks did not appear to involve Anthropic’s newer Mythos model. This is significant because the company had earlier warned that such advanced models could be weaponized by hackers in ways similar to what is now being observed. The release of Mythos had previously prompted global security teams to address thousands of vulnerabilities the model had identified, underscoring the dual nature of these powerful tools.

This incident is part of a broader pattern of AI being exploited by bad actors. Anthropic previously disclosed in November 2025 that its AI was used in a cyberattack. More recently, Claude was utilized to compromise the security of competitor OpenAI, though that specific incident was carried out by researchers rather than criminal groups. The recurrence of these events suggests that AI-powered threats are becoming a persistent and evolving challenge for the industry.

The timing of this revelation coincides with similar investigations in South Korea, where authorities are examining AI-related security issues affecting major banks including Hana Bank, KB Kookmin Bank, and Shinhan Bank. In that case, personal credit information for at least 25,000 Shinhan Bank customers was leaked, indicating that the threat is not confined to the Latin American region.

In response to these growing risks, technology giants are moving to secure the infrastructure. Nvidia recently launched the Open Agent Safety Platform, a hardware and software stack designed to prevent rogue AI agents from escaping their sandboxes or executing unauthorized code. The platform aims to mitigate the risk of agents gaining access to critical infrastructure or bypassing existing guardrails, a necessary step as AI becomes more central to global operations.

As AI agents become more autonomous, the line between helpful automation and automated attack continues to blur. For New England readers and businesses globally, the lesson is clear: the speed of cyberattacks is no longer limited by human reflexes, but by the processing power of the models driving them.