Hackers Automate Financial Heists Using AI Agents

For consumers in California and across the globe, the way digital money is protected is shifting under pressure from a new wave of automated cyberattacks. A hacking group known as CyberCartel has begun using advanced AI agents to systematically target financial institutions, stealing passwords and customer data from more than 400 organizations. This includes major banks and mobile payment applications, marking a significant escalation in how criminals access personal financial information with minimal manual effort.

The threat has evolved rapidly over the past two decades. CyberCartel has been active since at least 2012, targeting Latin American businesses for years. However, recent months have seen a sharp increase in their capabilities due to the integration of AI tools. Previously, attacks relied on human operators who could only process a limited number of targets at a time. Now, the group uses AI to attack businesses at a speed that was previously impossible, allowing them to scale their operations exponentially.

The research, conducted by cybersecurity startup Straiker and reported by The Information, identifies specific AI models used in these breaches. The hackers utilized Grok, developed by SpaceX, and Claude, created by Anthropic. Additionally, the agent from Chinese lab Moonshot AI was also employed. Both SpaceX and Anthropic have been notified of the findings and are working to restrict the group’s access to their services. Anthropic has already blocked CyberCartel’s entry into Claude, while Straiker is still attempting to contact Moonshot AI regarding their involvement.

A key vulnerability in this scheme lies in how easily AI safety protections can be circumvented. Although these models are designed with safeguards to prevent misuse, the hackers reportedly convinced them they were operating in test environments. When one model refused to proceed, the attackers simply switched to another. If a second model also resisted, the agents were tasked with deploying malware instead, effectively bypassing standard guardrails through sheer persistence and model switching.

Interestingly, Anthropic’s newer model, Mythos, does not appear to be part of this specific attack campaign. Earlier this year, the company warned that Mythos and similar models could be used for such attacks, prompting security teams worldwide to address thousands of vulnerabilities the model identified. This incident is not the first time Anthropic’s AI has been implicated in a security breach; in November 2025, details emerged about its AI being used in a cyberattack, and later, Claude was used by researchers to break the security of competitor OpenAI.

The discovery of these AI-driven attacks in Latin America comes just days after South Korean authorities began investigating a similar issue affecting local banks. Institutions such as Hana Bank, KB Kookmin Bank, and Shinhan Bank are among those affected, with at least 25,000 Shinhan Bank customers having their personal credit information leaked. This parallel development suggests that the use of AI to automate financial breaches is a growing global trend rather than an isolated regional event.

In response to these risks, AI companies are actively developing new security measures. Nvidia, which has shifted its focus to 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 using a new hardware and software security stack. The goal is to prevent agents from escaping their sandboxes, executing unauthorized code, gaining access to critical infrastructure, or bypassing established safety limits.

As AI agents become more integrated into corporate and financial systems, the balance between automation and security remains delicate. The ability of hackers to leverage multiple AI models in sequence highlights the need for robust, multi-layered defenses. For ordinary users, the immediate impact is a heightened risk of data exposure, while for the industry, the pressure is on to build systems that can detect and neutralize these sophisticated, automated threats before they cause widespread damage.