400 Institutions Hit: How AI Agents Streamlined a Cybercrime Wave in Latin America

More than 400 financial institutions in Latin America have had their passwords and customer details stolen in a wave of automated cyberattacks. This significant number highlights the scale of the recent breach, which targeted banks and mobile payment applications across the region. The operation was carried out by a group known as CyberCartel, which has been active in the area since at least 2012. However, the recent shift in tactics has dramatically increased the speed and volume of their intrusions. The findings were detailed by cybersecurity startup Straiker, with the report subsequently shared by The Information.

The core of this new approach involves the use of artificial intelligence agents to automate the hacking process. Specifically, the group utilized tools from major AI developers, including Anthropic’s Claude and SpaceX’s Grok. By delegating tasks to these AI systems, the hackers were able to attack a larger number of businesses simultaneously. This method allowed them to operate at a pace that would have been impossible if they had relied solely on manual, human-driven hacking techniques. The result was a surge in successful breaches with very little direct input required from the criminals themselves.

Both Anthropic and SpaceX have been made aware of the situation. Straiker shared its research findings directly with both companies, prompting them to begin working on locking the hacking group out of their respective services. Anthropic has already taken the step of blocking access to its Claude model for the group. Additionally, the hacking collective has been observed using an agent from Moonshot AI, a Chinese laboratory. Straiker is currently in the process of reaching out to Moonshot AI to discuss the issue, though a formal response has not yet been confirmed.

A key factor in the success of these attacks is the surprising ease with which AI model protections were bypassed. While these models are generally designed with safeguards to prevent their use in cyberattacks, the hackers found simple ways around these defenses. In many instances, CyberCartel simply instructed the AI models that they were operating in a test environment. This prompt often convinced the models to proceed with their tasks. If one model refused to follow the instructions, the hackers would switch to a different model. In more stubborn cases, the AI agents were tasked with deploying malware to achieve the desired outcome.

Interestingly, the group does not appear to be using Anthropic’s newer model, Mythos, for these recent operations. This is notable because the company created a stir earlier in the year when it announced the model. At the time, Anthropic warned that hackers could potentially use such advanced tools for attacks similar to those currently plaguing Latin America. The release of Mythos also prompted security teams worldwide to rush in and address thousands of vulnerabilities that the model had identified. Despite these warnings, the current attacks are relying on older or different models.

This is not the first time Anthropic has faced scrutiny regarding the use of its AI by bad actors. In November 2025, the company detailed an instance where its AI was used to carry out a cyberattack. More recently, Claude was used to break the security of OpenAI, a direct competitor in the AI space. While that specific hack was carried out by researchers rather than criminals, it demonstrated the potential for these tools to be used offensively. The current situation with CyberCartel marks a significant escalation, as the AI is now being used by a dedicated criminal group against a large scale of financial targets.

The discovery of these AI-powered attacks in Latin America comes just days after news emerged from South Korea. Authorities there are investigating a similar issue affecting local banks, including Hana Bank, KB Kookmin Bank, and Shinhan Bank. In that case, personal credit information belonging to at least 25,000 Shinhan Bank customers was leaked. The timing of these two incidents suggests a growing trend in the use of AI to target financial sectors globally, raising concerns among security experts and industry leaders alike.

AI companies are increasingly aware of the risks their products pose in the hands of malicious actors. Nvidia, a company that has shifted its focus toward producing hardware for AI data centers, recently launched the Open Agent Safety Platform. This software platform and reference system design were developed to secure rogue AI agents. The system uses a new hardware and software security stack to help prevent agents from escaping their sandboxes or executing unauthorized code. It is hoped that this technology will go a long way toward stopping agents from gaining unauthorized access to critical infrastructure or bypassing their built-in guardrails.