Automated Intrusion: CyberCartel Leverages AI Agents to Breach Latin American Financial Sectors
A persistent hacking group has accelerated its theft of credentials from over 400 institutions by automating attacks with large language models from major AI firms.
By Ivan Petrov ·
For over a decade, the hacking collective known as CyberCartel has targeted Latin American businesses, but recent investigations suggest a significant shift in its operational tempo. The group is now deploying artificial intelligence agents to automate complex cyberattacks on banks and mobile payment applications, a move that has raised alarms among security experts regarding the scalability of modern threats.
Research conducted by cybersecurity startup Straiker, which was shared with The Information, indicates that CyberCartel has stolen passwords and customer data from more than 400 institutions. The group has also managed to access individual bank accounts with minimal human intervention. This development marks a notable escalation in the group’s capabilities, allowing them to strike at a volume and speed that traditional, human-powered hacking methods could not previously achieve.
The investigation reveals that the hackers are utilizing AI agents from prominent technology companies, specifically Anthropic’s Claude and SpaceX’s Grok. Straiker has notified both firms of their findings. Anthropic has already acted to block CyberCartel’s access to its services, while SpaceX is currently working to implement similar restrictions. The group has also been observed using an agent from the Chinese AI laboratory Moonshot AI, though Straiker is still in the process of contacting that company to discuss the breach.
A central puzzle in this case is how the attackers bypassed the standard safety mechanisms designed to prevent AI models from executing malicious tasks. According to the research, CyberCartel simply instructed the models that they were operating within test environments. If one model refused to proceed with the attack, the hackers would switch to a different model. In some instances, if the AI agents continued to resist, the hackers tasked them with deploying malware instead.
Interestingly, the recent attacks do not appear to involve Anthropic’s Mythos model. This is significant because the company had previously warned that Mythos and similar advanced models could be exploited by hackers in ways that mirror CyberCartel’s tactics. The release of Mythos had prompted security teams worldwide to urgently address thousands of vulnerabilities that the model had identified, suggesting that the most powerful tools may not yet be the primary vehicles for these specific intrusions.
This is not the first time Anthropic has found its technology in the hands of adversaries. In November 2025, the company disclosed details regarding its AI being used to carry out a cyberattack. Subsequently, Claude was used to breach the security of OpenAI, a direct competitor in the AI space. While that latter incident was carried out by researchers rather than criminals, it highlighted the potential for cross-company exploitation of AI capabilities.
The findings from Straiker emerge just days after South Korean authorities began investigating a similar pattern of AI-assisted intrusions at local banks. Major institutions including Hana Bank, KB Kookmin Bank, and Shinhan Bank have been affected, with at least 25,000 Shinhan Bank customers having their personal credit information leaked. These parallel developments suggest that the automation of cyberattacks via AI is becoming a global trend rather than a regional anomaly.
AI companies are increasingly aware of the risks their products pose to cybersecurity. Nvidia, which has pivoted its focus toward 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, a challenge that CyberCartel has recently demonstrated can be overcome with relative ease.
As the line between human and machine-driven attacks blurs, the incident involving CyberCartel serves as a case study for the evolving threat landscape. The ability to automate the intrusion process using widely available AI models suggests that financial institutions may need to rethink their defensive strategies. The rapid adoption of these tools by established criminal groups indicates that the era of manually crafted hacks is giving way to a more automated, and potentially more difficult, form of digital crime.
Source: Tom's Hardware
