CyberCartel Expands Latin American Campaign Using AI Agents
A veteran hacking group has automated its attacks on regional banks and payment apps by leveraging AI models from major tech firms.
By Robert Hayes ·
The automation of cybercrime has reached a new scale as a prominent hacking group integrates artificial intelligence into its operations. This shift marks a significant departure from traditional manual methods, allowing for a surge in the volume and speed of attacks.
The group known as CyberCartel has targeted Latin American businesses since at least 2012. However, recent months have seen a distinct escalation in their capabilities. They are now utilizing AI agents from SpaceX and Anthropic to streamline their intrusions. This technological upgrade enables the gang to compromise more targets simultaneously than was previously possible with human-led efforts.
Research from the cybersecurity startup Straiker details this evolution. The findings indicate that the group has stolen passwords and customer data from over 400 institutions. These targets include major banks and mobile payment applications. The level of automation means that criminals require minimal input to access individual bank accounts. Straiker shared these insights with The Information, highlighting the growing threat of AI-driven cyberattacks.
Both SpaceX and Anthropic have been informed of the group's activities. The two companies are currently working to prevent CyberCartel from accessing their services. Anthropic has already succeeded in blocking access to its Claude model. The hacking group has also utilized an agent from the Chinese lab Moonshot AI. Straiker is in the process of contacting that company to discuss the matter.
The effectiveness of these attacks stems from the ease with which AI safeguards can be circumvented. Although models typically include protections against cyber use, CyberCartel reportedly bypassed them by claiming the tools were in test environments. If one model refused to proceed, the hackers switched to a different one. In some cases, the agents were tasked with deploying malware directly. This strategy highlights a potential vulnerability in current AI safety protocols.
Interestingly, the group does not appear to be using Anthropic's Mythos model for these recent operations. That model generated significant attention earlier this year due to warnings about its offensive potential. Its release prompted security teams worldwide to address thousands of newly identified vulnerabilities. The current campaign relies instead on the more established Claude and Grok systems.
This incident is not the first time Anthropic has dealt with its technology being used by malicious actors. In November 2025, the company revealed that its AI had been used to execute a cyberattack. Subsequently, Claude was employed to breach the security of OpenAI, a direct competitor. While that specific hack was carried out by researchers rather than criminals, it demonstrated the model's offensive capabilities.
The findings from Straiker arrive shortly after reports of similar issues in South Korea. Local authorities are investigating attacks that affected several major banks, including Hana Bank and KB Kookmin Bank. At least 25,000 customers of Shinhan Bank had their personal credit information leaked. This parallel suggests a global trend in the use of automated tools against financial sectors.
The broader industry is taking note of these risks. Nvidia recently launched the Open Agent Safety Platform to address these concerns. This software and hardware stack is designed to secure rogue AI agents. It aims to prevent agents from escaping their designated sandboxes or executing unauthorized code. The platform is intended to stop agents from gaining access to critical infrastructure or bypassing existing guardrails.
As AI agents become more central to digital infrastructure, the balance between utility and security remains a critical challenge. The case of CyberCartel illustrates how quickly established threats can adapt to new technologies. For financial institutions in the Midwest and beyond, this serves as a reminder that the perimeter of defense must expand to include the automated tools their adversaries are now employing.
Source: Tom's Hardware
