CyberCartel Automation: AI Agents Drive Mass Breaches of Latin American Finance

CyberCartel, a hacking collective active since 2012, has fundamentally altered its operational tempo by integrating AI agents into its attack infrastructure. The group utilized autonomous software tools, specifically Grok from SpaceX and Claude from Anthropic, to automate intrusions into the financial sector. This shift has enabled the group to breach more than 400 institutions across Latin America, including major banks and mobile payment platforms. The result is a significant exfiltration of customer data, including passwords and sensitive personal details, with minimal manual intervention required from the human operators behind the scenes.

The findings, developed by the cybersecurity startup Straiker and shared by The Information, highlight a new era of AI-driven cybercrime. For years, CyberCartel relied on traditional, human-powered hacking methods. However, recent months have seen a dramatic increase in the speed and volume of its attacks. By offloading repetitive tasks to AI agents, the group can now target a larger number of businesses simultaneously. This capability marks a significant escalation in threat levels for financial institutions in the region.

The use of commercial AI models for malicious purposes has prompted immediate responses from the technology providers. Straiker shared its research with both SpaceX and Anthropic to facilitate mitigation efforts. Anthropic has already taken steps to block CyberCartel’s access to its Claude service. SpaceX is also working to secure its Grok platform against the group. Additionally, the hacking organization has been observed using an agent from Moonshot AI, a Chinese laboratory. Straiker is currently attempting to establish contact with Moonshot AI to coordinate further defensive measures.

Despite the fact that most AI models are designed with built-in safeguards to prevent their use in cyberattacks, these protections proved easy for CyberCartel to circumvent. The group reportedly deceived the models by claiming the software was operating in a test environment. This simple prompt manipulation allowed the agents to bypass standard safety checks. If one model refused to execute a specific task, the hackers would switch to a different model. In some instances, the AI agents were instructed to deploy malware directly if their initial attempts were blocked.

Interestingly, the group did not appear to rely on Anthropic’s Mythos model for these recent operations. Anthropic had previously generated significant industry attention when it released Mythos, warning that advanced models could be exploited for attacks similar to those now being observed. The release of Mythos had also prompted security teams worldwide to address thousands of vulnerabilities identified by the model. However, the current wave of attacks by CyberCartel relies on the more widely available Claude and Grok systems.

This incident is not the first time Anthropic has addressed the use of its technology by malicious actors. In November 2025, the company detailed how its AI was utilized to carry out a cyberattack. Subsequently, Claude was employed by researchers to compromise the security of OpenAI, a direct competitor in the AI market. While that specific breach was conducted by academic researchers rather than criminal gangs, it underscored the potential for AI to be used as an offensive tool in the broader technology landscape.

The timing of this research is notable, as it arrives just days after South Korean authorities began investigating similar issues affecting domestic banks. Hana Bank, KB Kookmin Bank, and Shinhan Bank are among the institutions under scrutiny in that region. The investigation has revealed that personal credit information belonging to at least 25,000 Shinhan Bank customers was leaked. These parallel developments suggest that AI-assisted attacks on financial data are becoming a global phenomenon rather than a regional one.

Major technology firms are increasingly aware of the risks associated with deploying autonomous agents. Nvidia, which has shifted its focus toward producing hardware for AI data centers, recently launched the Open Agent Safety Platform. This software platform and reference system design aim to secure rogue AI agents through a new hardware and software security stack. The system is intended to prevent agents from escaping their designated sandboxes, executing unauthorized code, or bypassing established guardrails to access critical infrastructure. These developments reflect a growing industry consensus that robust security measures are essential for the safe deployment of AI agents in sensitive environments.