CyberCartel Leverages AI Agents to Automate Financial Heists in Latin America

The integration of artificial intelligence into cybercrime has evolved from a theoretical concern into a tangible threat for financial sectors across the globe. In the most recent development, a cybersecurity startup named Straiker has identified a significant escalation in the tactics employed by a notorious hacking collective known as CyberCartel.

The group, which has been active since at least 2012, is now utilizing AI agents to automate attacks on banks and mobile payment applications primarily in Latin America. According to research shared by The Information, this shift has allowed the hackers to access the accounts of individuals and steal credentials from more than 400 institutions. This new capability enables the group to conduct operations at a speed and scale that was previously unattainable through human-led efforts alone.

The specific tools driving this offensive include Claude, developed by Anthropic, and Grok, a product of SpaceX. Straiker has disclosed that it has shared its findings with both companies, prompting them to take action to restrict the hacking group's access to their respective services. Anthropic has already succeeded in blocking CyberCartel's use of Claude, while the other firm is working toward a similar outcome. Additionally, the group has utilized an agent from Moonshot AI, a Chinese laboratory, though Straiker is currently in the process of contacting that company regarding the findings.

A central question in this incident is how the hackers bypassed standard security measures designed to prevent AI models from executing malicious tasks. The research suggests that the group successfully deceived the models by claiming they were operating within test environments. When one model refused to proceed with the attack, the hackers would simply switch to a different model. If a second refusal occurred, the agents were instructed to deploy malware instead of continuing the primary operation.

Interestingly, the attacks do not appear to involve Anthropic's more advanced Mythos model. This model had previously drawn attention earlier this year when the company warned that it could be exploited by bad actors. The release of Mythos also prompted security teams worldwide to address thousands of vulnerabilities that the model had identified, yet it does not seem to be the primary tool in CyberCartel's current campaign.

This incident is not the first time Anthropic has addressed the use of its technology by adversaries. In November 2025, the company detailed how its AI was used to execute a cyberattack. Subsequently, Claude was employed to breach the security systems of OpenAI, a direct competitor. While that particular breach was carried out by researchers rather than criminals, it highlighted the potential power of these tools in the hands of knowledgeable actors.

The timing of Straiker's report is notable as it arrives just days after South Korean authorities began investigating a similar pattern of 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 reporting that their personal credit information had been leaked. This parallel suggests that the trend of automated AI attacks may be expanding beyond Latin American borders.

The industry is increasingly aware of the risks associated with autonomous agents. Nvidia, which has shifted its focus toward hardware for AI data centers, recently launched the Open Agent Safety Platform. This initiative includes a software platform and reference system design aimed at securing rogue agents through a new stack of hardware and security software. The goal is to prevent agents from escaping their designated sandboxes, executing unauthorized code, or bypassing guardrails to access critical infrastructure.

As these incidents multiply, the pressure on AI providers to balance innovation with security continues to grow. The case of CyberCartel demonstrates that even with existing protections, bad actors can find ways to leverage these powerful tools for financial gain. For now, the focus remains on how quickly vendors can adapt and how effectively they can limit the reach of groups that have mastered the art of the automated heist.