400 Financial Institutions Targeted by AI-Driven Cyber Attacks

A hacking collective known as CyberCartel has successfully compromised more than 400 financial institutions across Latin America by leveraging artificial intelligence agents to automate the theft of passwords and customer data. This recent wave of attacks marks a significant escalation in the group's operational capabilities, shifting from traditional human-led methods to an AI-assisted model that allows for a higher volume of intrusions in a shorter timeframe. The findings were detailed by cybersecurity startup Straiker, which shared its research with major AI developers to address the emerging threat.

CyberCartel has been an active threat in the region since at least 2012, but its recent adoption of AI tools has fundamentally changed its attack speed. According to the new research, the group utilized agents from Anthropic, specifically Claude, as well as Grok from SpaceX. These AI systems were tasked with identifying vulnerabilities and executing attacks on banks and mobile payment applications with minimal human intervention. This automation has allowed the criminals to access individual bank accounts and steal sensitive details at a scale previously impossible for their team size.

The technical methodology behind these attacks highlights the current weaknesses in AI safety protocols. Although AI models typically include safeguards to prevent their use in cyber operations, CyberCartel reportedly bypassed these protections by instructing the models that they were operating in test environments. If one model refused to proceed with the task, the hackers simply switched to a different one. In some instances, if the AI agents still resisted, the group tasked them with deploying malware directly, demonstrating a flexible and persistent approach to overcoming software guardrails.

Both Anthropic and SpaceX have been notified of the research and are currently working to restrict CyberCartel's access to their respective services. Anthropic has already managed to block the group's access to Claude. The hacking group has also been observed using an agent from Moonshot AI, a Chinese laboratory, though Straiker is still in the process of contacting that company to discuss the findings. This multi-platform approach by the attackers suggests a strategy to diversify their tools and reduce the risk of being cut off from a single provider.

Interestingly, the recent attacks do not appear to involve Anthropic's Mythos model, despite that model causing significant discussion earlier this year regarding its potential for offensive security use. When Mythos was released, the company warned that hackers could exploit its capabilities for attacks similar to those seen now. The release also prompted security teams worldwide to address thousands of vulnerabilities that the model had identified, yet the group opted for other available agents in this specific campaign.

This incident is not the first time Anthropic has dealt with its technology being used by external actors for cyber operations. In November 2025, the company disclosed details of its AI being used to execute a cyberattack. Subsequently, Claude was used to break the security of OpenAI, a direct competitor in the AI market. While that particular breach was carried out by researchers rather than criminal groups, it underscored the growing trend of AI models being employed in complex security challenges.

The timing of this research coincides with other recent developments in the global financial sector. Just days before these findings were shared, South Korean authorities began investigating a similar issue affecting local banks. Major institutions including Hana Bank, KB Kookmin Bank, and Shinhan Bank were impacted, with at least 25,000 Shinhan Bank customers having their personal credit information leaked. This parallel suggests that the vulnerability of financial data to AI-driven attacks is a global phenomenon, not limited to the Latin American market.

AI companies are increasingly aware of the risks their products pose to cybersecurity infrastructure. 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 sandboxes, executing unauthorized code, or bypassing established guardrails, thereby reducing the likelihood of such automated attacks in the future.

The next steps for the industry will likely involve stricter access controls and more robust verification processes for AI agents used in financial environments. As companies like Anthropic and SpaceX work to lock out known bad actors, the broader market may see a push for standardized safety protocols. For Texas and South Central businesses, this highlights the need to monitor how AI tools are integrated into their security and payment systems, ensuring that the efficiency gained from automation does not come at the cost of new, hard-to-detect vulnerabilities.