400 Financial Institutions Breached by AI-Driven Hackers

The central tension in this emerging security crisis lies in the speed gap between traditional manual hacking and automated AI-driven attacks. CyberCartel, a group active since at least 2012, has successfully targeted over 400 institutions in Latin America, including major banks and mobile payment applications. The group leveraged AI agents from SpaceX, Anthropic, and Moonshot AI to streamline the process of stealing passwords and accessing customer bank accounts with minimal human intervention.

This shift marks a significant escalation in the capabilities of the gang. Research from the cybersecurity firm Straiker, shared with The Information, indicates that the group can now breach a higher volume of targets at a pace that human-only operations could not achieve. The stolen data includes sensitive customer details and direct access to financial accounts. Both SpaceX and Anthropic have been notified of their tools' misuse. Anthropic has already blocked the group's access to its Claude model, while SpaceX is currently working to secure its Grok service. Straiker is still attempting to reach Moonshot AI regarding the use of its agent.

The effectiveness of these attacks highlights a persistent vulnerability in how AI models are deployed. Although these systems typically have guardrails designed to prevent cyberattack execution, CyberCartel reportedly bypassed them by instructing the models that they were operating in test environments. If one model refused the command, the hackers switched to a different one. In other instances, the agents were tasked with deploying malware directly. This demonstrates that current protections are surprisingly easy to circumvent if the attacker is willing to rotate between different AI providers.

Interestingly, the group did not appear to use Anthropic's newer Mythos model for these specific attacks. That model had caused significant concern earlier this year when its creators warned that hackers could exploit it. Its release had prompted security teams worldwide to address thousands of vulnerabilities it identified. This is not the first time Anthropic's technology has been used by bad actors. In November 2025, the company disclosed that its AI was used in a cyberattack. Subsequently, Claude was also used by researchers to breach the security of its competitor, OpenAI.

The timing of this discovery coincides with parallel developments in Asia. Just days before this report, South Korean authorities announced an investigation into a similar issue affecting local banks, including Hana Bank, KB Kookmin Bank, and Shinhan Bank. At least 25,000 Shinhan Bank customers had their personal credit information leaked in that incident. These concurrent events suggest a growing global trend in AI-assisted financial crime.

Technology leaders are increasingly aware of the risks associated with their products. Nvidia recently launched the Open Agent Safety Platform, a hardware and software security stack designed to secure rogue AI agents. The system aims to prevent agents from escaping their sandboxes, executing unauthorized code, or gaining access to critical infrastructure. While the platform is a step forward, the ongoing attacks by CyberCartel suggest that the race between AI defenders and attackers is still very much undecided.