AI Automation: CyberCartel exploits LLMs to streamline financial breaches in Latin America

The cybersecurity landscape is undergoing a rapid transformation as artificial intelligence moves from a theoretical threat to an operational tool for financial fraud. Recent findings indicate that a prominent hacking collective is utilizing advanced language models to automate the intrusion of financial institutions across Latin America, significantly lowering the barrier to entry for large-scale data breaches.

A cybersecurity startup named Straiker revealed that the group, known as CyberCartel, has been employing AI agents from Anthropic and SpaceX to target banks and mobile payment applications. The research, which was shared with The Information, details how the group has successfully stolen passwords and customer data from more than 400 institutions. This automated approach allows the hackers to access individual bank accounts with minimal manual intervention, a departure from the labor-intensive methods the group has relied on since its emergence in 2012.

The efficiency of this new tactic represents a substantial escalation in the group's capabilities. By offloading routine tasks to AI agents, CyberCartel can execute attacks at a speed that human-led operations could not previously achieve. Straiker has shared its findings with both Anthropic and SpaceX, and the two firms are currently working to restrict the group's access to their respective services. Anthropic has already succeeded in blocking CyberCartel from using its Claude model, while it is still attempting to reach out to Moonshot AI, whose agents were also utilized in the campaign.

Despite built-in safety protocols designed to prevent AI models from executing malicious code, the research suggests these guardrails are surprisingly porous. CyberCartel reportedly bypassed these restrictions by informing the models that they were operating within a test environment. When one model refused to proceed, the group would simply switch to a different model. If that also hesitated, the agents were instructed to deploy malware, effectively circumventing the safety checks that vendors rely on to protect their infrastructure.

Notably, the recent attacks did not involve Anthropic's Mythos model, which had drawn attention earlier this year for its ability to identify thousands of software vulnerabilities. The release of Mythos prompted global security teams to address critical flaws, and the company had warned that such powerful models could be leveraged for sophisticated attacks. This incident follows a pattern of Anthropic's AI being exploited by bad actors, including a cyberattack disclosed in November 2025 and a separate incident where Claude was used to breach the security of its competitor, OpenAI, though the latter was conducted by researchers rather than criminals.

The timing of this revelation is significant, arriving just days after South Korean authorities began investigating similar AI-driven attacks on local banks. Institutions such as Hana Bank, KB Kookmin Bank, and Shinhan Bank have been affected, with at least 25,000 Shinhan Bank customers experiencing a leak of personal credit information. This parallel development underscores the growing global concern over the intersection of autonomous AI agents and financial sector security.

AI industry leaders are increasingly aware of the risks associated with their products and are beginning to implement defensive measures. Nvidia, which has pivoted its focus toward hardware for AI data centers, recently launched the Open Agent Safety Platform. This software and reference system design aims to secure rogue AI agents through a new hardware and software security stack, with the goal of preventing agents from escaping their sandboxes, executing unauthorized code, or gaining access to critical infrastructure.