AI-Powered Hackers Breach Hundreds of Latin American Banks

For residents in Latin America who rely on digital banking, the landscape of personal security has shifted dramatically. A prominent hacking collective known as CyberCartel has begun automating its intrusions using powerful artificial intelligence agents, resulting in the theft of passwords and customer data from more than 400 institutions. These targets include major banks and popular mobile payment applications, meaning that individuals' bank accounts are now being accessed with significantly less manual effort from the criminals. This development marks a pivotal moment where AI-driven tools are being deployed at scale against the financial infrastructure of the region.

The findings come from a cybersecurity startup named Straiker, whose research was highlighted by The Information. The study reveals that CyberCartel, a threat actor active since at least 2012, has recently integrated AI models to accelerate its operations. By leveraging these automated systems, the group can now strike at a speed and volume that was previously unattainable with human-led hacking efforts. The primary AI tools identified in these attacks include Grok, developed by SpaceX, and Claude, created by Anthropic. Additionally, the group has utilized an agent from Chinese lab Moonshot AI, though Straiker is still in the process of contacting that company regarding the findings.

Both SpaceX and Anthropic have been notified of the research and are currently working to restrict CyberCartel’s access to their respective platforms. Anthropic has already succeeded in blocking the group’s use of Claude, a move aimed at cutting off one of the key resources powering the automated breaches. The speed of this response underscores the urgency with which AI providers are viewing the misuse of their models. Meanwhile, the continued use of the Moonshot AI agent highlights a broader trend in the global AI market, where multiple vendors' technologies are converging in the hands of cybercriminals.

The mechanics of these attacks reveal a surprising vulnerability in how AI models are protected. While most AI systems have built-in safeguards to prevent their use in cyber operations, CyberCartel reportedly bypassed these defenses by simply instructing the models that they were operating in a test environment. When one model refused to proceed with the task, the hackers would switch to a different model. If that alternative also resisted, the AI agents were tasked with deploying malware directly. This tactic of shifting between models allows the group to maintain a steady stream of automated attacks despite individual model restrictions.

Notably, the recent wave of attacks does not appear to involve Anthropic’s newer model, Mythos. This model had caused significant discussion earlier in the year when the company warned that such advanced tools could be exploited by hackers in ways similar to those currently seen. The release of Mythos had previously prompted security teams worldwide to address thousands of vulnerabilities that the model identified, suggesting that while the latest tools are powerful, they are not the only ones being weaponized by criminal groups.

This is not the first time Anthropic has dealt with the implications of its AI being used by external actors. In November 2025, the company disclosed details of its AI being employed in a cyberattack. Subsequently, Claude was used to breach the security systems of OpenAI, a direct competitor in the AI sector. Although that specific breach was carried out by researchers rather than criminals, it demonstrated the potential for cross-organizational exploitation of AI capabilities. The current situation with CyberCartel represents a more direct threat to consumer-facing financial services.

The timing of these findings is particularly significant as South Korean authorities launch their own investigations into similar issues affecting local banks. Institutions including Hana Bank, KB Kookmin Bank, and Shinhan Bank are under scrutiny, with reports indicating that personal credit information for at least 25,000 Shinhan Bank customers has been leaked. This parallel development suggests that the use of AI in financial cybercrimes is a global phenomenon, not confined to the Latin American market.

In response to these emerging risks, AI hardware and software companies are moving to fortify their defenses. Nvidia, which has shifted its focus toward producing hardware for AI data centers, recently launched the Open Agent Safety Platform. This system is designed to secure rogue AI agents through a new stack of hardware and software security measures. The platform aims to prevent agents from escaping their designated sandboxes, executing unauthorized code, or gaining access to critical infrastructure, offering a potential solution to the growing problem of autonomous cyberattacks.

As the integration of AI into cybercriminal tactics accelerates, the burden on financial institutions and AI providers grows. For ordinary users, the implication is that the defenses protecting their digital assets must evolve as quickly as the tools being used to break them. The next phase of this conflict will likely see further testing of these new safety platforms and continued efforts by AI companies to lock out known bad actors from their services.