AI Agents Enable CyberCartel to Breach Hundreds of Latin American Financial Entities

The rapid adoption of artificial intelligence by cybercriminals has transformed the threat landscape for financial institutions across the Americas, allowing a single group to execute attacks at a scale previously impossible for human teams.

A new investigation by cybersecurity startup Straiker, reported by The Information, reveals that the hacking collective known as CyberCartel has been leveraging AI agents to automate intrusions into Latin American banks and mobile payment applications. Since at least 2012, the group has targeted the region, but recent months have seen a significant escalation in capability and speed. By integrating AI tools, CyberCartel has been able to compromise more than 400 institutions, extracting passwords and customer data while accessing individual bank accounts with minimal manual intervention from the criminals themselves.

The automation of these attacks relies heavily on specific large language models, notably Grok from SpaceX and Claude from Anthropic. These AI agents facilitate the technical steps of the intrusion, allowing the group to process data and navigate security systems far faster than traditional human-led methods. Straiker has shared its findings with both technology giants, prompting immediate action. Anthropic has already blocked the group’s access to its Claude model, while SpaceX is currently working to exclude CyberCartel from its services. The group has also utilized agents from Chinese lab Moonshot AI, though Straiker is still attempting to establish contact with that company to discuss the incident.

A critical vulnerability in the current deployment of AI agents lies in their susceptibility to social engineering, a technique that has proven surprisingly effective against modern language models. Although these systems typically include safeguards to prevent their use in cyberattacks, CyberCartel reportedly bypassed these protections by simply instructing the models that they were operating in a test environment. When one model refused to proceed, the hackers would switch to a different model. If that alternative also resisted, the agents were tasked with deploying malware, demonstrating a flexible and persistent approach to overcoming automated guardrails.

Notably, the group does not appear to be utilizing Anthropic’s newer Mythos model for these recent operations. The model had generated significant attention earlier in the year when the company warned that adversaries could exploit it for similar attacks. Its release also prompted security teams worldwide to address thousands of vulnerabilities that the model had identified. This incident marks another instance where Anthropic’s technology has been exploited by bad actors, following a November 2025 disclosure of an AI-led cyberattack and a later case where Claude was used by researchers to breach the security of competitor OpenAI.

The findings from Straiker arrive just days after South Korean authorities announced they were investigating a similar surge in AI-driven attacks on local banks. Major institutions, including Hana Bank, KB Kookmin Bank, and Shinhan Bank, have been affected, with at least 25,000 Shinhan Bank customers reporting leaks of personal credit information. This parallel development highlights a global trend in which financial sectors are becoming primary targets for AI-augmented threats.

In response to these emerging risks, major technology firms are developing new layers of defense. Nvidia, which now focuses heavily on 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 through a new hardware and software security stack. The goal is to prevent agents from escaping their designated sandboxes, executing unauthorized code, or gaining access to critical infrastructure, offering a potential pathway to mitigate the growing dangers of AI-powered cybercrime.