Hackers Automate Bank Intrusions With AI Agents

At least 25,000 customer records were leaked from Shinhan Bank as South Korean authorities investigate a wave of AI-driven attacks on the nation's financial sector. The breach follows new research from cybersecurity startup Straiker, which details how the hacking group CyberCartel has deployed AI agents to automate thefts from Latin American banks and payment apps.

CyberCartel has operated since 2012, but Straiker found that the integration of AI tools has allowed the group to target more organizations at a speed that human-led attacks cannot match. The group has stolen passwords and customer data, granting access to individual bank accounts with minimal direct involvement from the criminals. Straiker identified agents from SpaceX, known as Grok, and from Anthropic, known as Claude, as primary tools in the operation. An agent from the Chinese lab Moonshot AI was also utilized. Anthropic has blocked CyberCartel's access to Claude, and SpaceX is working to cut off the group's use of Grok. Straiker continues to seek contact with Moonshot AI.

The attacks exploit a weakness in how AI models handle safety instructions. These systems are designed to refuse requests that aid in cyberattacks. CyberCartel reportedly bypassed these safeguards by instructing the models that they were operating in test environments. When one model refused to continue, the hackers switched to another. If a second model also declined, the agents were assigned to deploy malware.

This incident is part of a larger pattern of AI models being used by adversaries. Anthropic disclosed in November 2025 that its AI was used in a cyberattack. The company also noted that Claude was later used to break the security of OpenAI, a rival developer, though that specific operation was conducted by researchers. Earlier this year, Anthropic warned that hackers could exploit its newly unveiled Mythos model for similar attacks. The release of Mythos prompted security teams worldwide to address thousands of vulnerabilities identified by the model. Straiker's report notes that CyberCartel does not appear to have used Mythos.

The South Korean investigation involves Hana Bank, KB Kookmin Bank, and Shinhan Bank. The leak of personal credit information from Shinhan Bank highlights the scale of the data exposed. The timing of the Straiker report, released on October 23, 2026, aligns closely with the public disclosure of the Korean bank breaches.

Industry leaders are developing defenses against these autonomous threats. Nvidia recently launched the Open Agent Safety Platform, a software and hardware system designed to contain rogue AI agents. The platform aims to prevent agents from escaping sandboxes, executing unauthorized code, or bypassing established security guardrails. The release provides a reference design for companies seeking to secure their infrastructure against AI-driven intrusions.