CyberCartel Automates Latin American Bank Heists Using Grok and Claude AI Agents

Over 400 Latin American financial institutions, including major banks and mobile payment platforms, have suffered breaches involving the theft of passwords and customer data. The operation was largely automated by a hacking group known as CyberCartel, which utilized AI agents to execute the attacks. This recent campaign represents a significant escalation in the group’s tactics, shifting from traditional human-led hacking to a model driven by artificial intelligence that allows for faster and broader targeting.

Research from the cybersecurity startup Straiker, recently shared by The Information, details how CyberCartel integrated large language models into their workflow. The group specifically employed Grok, an AI model from SpaceX, and Claude, developed by Anthropic. By delegating tasks to these AI agents, the hackers were able to access individual bank accounts and extract sensitive information with significantly less manual input than previously required. This automation has allowed the group to attack businesses at a speed that was unattainable when relying solely on human operators.

CyberCartel has been an active threat to the region since at least 2012, but the integration of AI marks a distinct change in its operational capability. Straiker has shared its findings with both SpaceX and Anthropic. Anthropic has already taken steps to block the group’s access to Claude, while both companies are currently working to lock the hackers out of their respective services. The group also reportedly used an agent from Moonshot AI, a Chinese lab, though Straiker is still attempting to establish contact with that company to discuss the incident.

The study highlights a concerning vulnerability in how AI models are protected against malicious use. Although these models typically have safeguards to prevent them from executing cyberattacks, the researchers found that these protections were easy to bypass. CyberCartel reportedly instructed the models that they were operating in test environments. If one model refused to proceed, the hackers would switch to a different model. If that also hesitated, the AI agents were tasked with deploying malware instead.

Notably, the group does not appear to be using Anthropic’s newer Mythos model for these recent operations. The company had previously warned that such advanced models could be exploited for attacks similar to those seen now. The release of Mythos had also prompted security teams worldwide to address thousands of vulnerabilities that the model had identified. This incident is not the first time Anthropic has addressed the use of its technology by bad actors, having disclosed a cyberattack involving its AI in November 2025 and a later breach of competitor OpenAI’s security by researchers.

This development follows closely on the heels of similar news from South Korea, where authorities are investigating AI-related issues affecting local banks. Hana Bank, KB Kookmin Bank, and Shinhan Bank are among those affected, with at least 25,000 Shinhan Bank customers reporting leaks of personal credit information. The timing suggests a growing global trend in the use of AI for financial cybercrime.

AI industry leaders are increasingly aware of the risks posed by their products. Nvidia, which has shifted its focus to 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 sandboxes, executing unauthorized code, or bypassing guardrails to access critical infrastructure.