Hacker's AI conversations exposed

A hacker suspected of attacking South Korean banks using artificial intelligence left his interactions with the AI system Claude accessible online. This allowed security researchers from CrowdStrike to piece together the hacker’s activities and methods.

South Korean banks targeted by cyberattacks

Since late September, several South Korean financial institutions have been victims of cyber intrusions that resulted in data theft. In one bank, a system for tracking loan requests meant for brokers was compromised, while another had its mobile work system accessed by the attacker.

AI tool ARTEX used in attacks

CrowdStrike researchers discovered the use of ARTEX, a tool that automates intrusion tests using artificial intelligence models. Originally developed in China and released under an open-source license, ARTEX allows AI to perform tasks usually done by cybersecurity experts, such as finding vulnerabilities and verifying their exploitability.

Multiple AI models used in the attack

The hacker primarily used DeepSeek v4.1-flash to run ARTEX. He also used Claude Code, an AI-powered environment for performing computer tasks, and two other models, GLM-5.3 and Grok 4.6, during additional sessions. These models helped automate various aspects of the cyberattack.

Evidence of sustained attacks found

Files analyzed by CrowdStrike show that the hacker was actively targeting multiple financial institutions between late September and early October. Researchers traced the attacks back to two servers controlled by the hacker, one hosting ARTEX and the other in Hong Kong, which acted as a central hub for his operations.

Personal details shared with AI

Among the files, researchers found ARTEX configurations and instructions written in Chinese, as well as the histories of Claude Code sessions from the cybercriminal. The hacker had shared personal information with the AI, including a name, phone number, Telegram identifier, university details, and a location in China’s Guangdong province.

AI assistance deemed critical

Adam Meyers, head of threat intelligence operations at CrowdStrike, told Reuters that the attacker likely couldn’t have executed the campaign without AI assistance. The hacker’s technical skills were limited, but the AI tools enabled him to target multiple institutions quickly.

ARTEX developer halts public access

Following the investigation, the developer of ARTEX announced the end of public distribution and maintenance of the software, condemning its use for illegal purposes. The project was removed from GitHub, though existing copies of the software continue to circulate.