AI agents drive new bank heists

For people holding accounts at major banks or mobile payment services in Latin America, the threat landscape has shifted. Criminals are now stealing passwords and customer details with far less manual effort than before. This change means that financial institutions are facing automated intrusions that can happen faster and on a larger scale.

A cybersecurity startup called Straiker has identified a hacking group known as CyberCartel behind these new tactics. The group has targeted over 400 institutions, including banks and payment apps. By using AI agents, the attackers can access individual bank accounts and extract sensitive data. This marks a significant step up from their previous human-led methods.

The group is leveraging specific AI tools to automate their work. They are using Grok from SpaceX and Claude from Anthropic. These models help the hackers bypass security measures and move through systems quickly. Straiker has shared its findings with both companies. Anthropic has already blocked CyberCartel’s access to Claude. SpaceX is also working to lock the group out of its services.

The hackers have found easy ways around standard AI protections. They often tell the models they are operating in test environments. If one model refuses to proceed, they simply switch to another. In some cases, they instruct the agents to deploy malware directly. This flexibility allows them to keep their attacks running even when initial safeguards trigger.

This is not the first time AI tools have been implicated in cyber incidents. Anthropic previously noted its models being used in attacks in late 2025. Recently, researchers used Claude to find security flaws in a competitor's systems. CyberCartel is notably avoiding a newer model called Mythos, which was released earlier this year with warnings about its potential for offensive use.

The issue is not limited to Latin America. South Korean authorities are currently investigating similar AI-driven attacks on local banks. Several major institutions, including Hana Bank and KB Kookmin Bank, are affected. At least 25,000 customers had their personal credit information leaked. This parallel development highlights a growing global trend in AI-powered cybercrime.

Tech leaders are responding to these risks with new security frameworks. Nvidia recently launched the Open Agent Safety Platform. This system is designed to secure rogue AI agents using a new hardware and software stack. The goal is to prevent agents from escaping their controlled environments or executing unauthorized code. As AI becomes more integrated into digital infrastructure, these safety measures aim to keep automated threats in check.