A hacker used an AI break-in agent on South Korea’s banks, then left the AI logs on an open server

Note

CrowdStrike Intelligence published a report on Oct 7 (US time) on a campaign against South Korean financial organizations that ran from late September to early October 2026 and ended with stolen data. CrowdStrike has not tied it to any named group or state. It assesses, with moderate confidence, that the attacker is a Chinese speaker motivated by money, based on the Chinese-built tool and Chinese-language prompts. A Financial Security Institute official told The Herald Business that AI was used but did not act on its own: “A hacker used the AI as a tool.”

The institute traced attack IPs and server logs at Shinhan Bank, the first to report a breach, and told The Herald Business it found clear evidence that ARTEX was used. Every attack hit internal employee or partner-facing systems, not customer internet or mobile banking. Shinhan says 25,729 people had names, phone numbers, annual income and loan limits exposed through a loan inquiry service used by brokers. KB Kookmin reported 119 affected customers, Hana Bank 89 records and BNK Busan Bank 11 contract workers. Police and financial regulators are investigating.

CrowdStrike describes ARTEX as a recently released open-source agentic penetration-testing tool developed in China: software that uses a large language model to probe systems for weaknesses and act on them. It found an ARTEX instance on one attacker-controlled server and, on a second server in Hong Kong, open directories holding Claude Code session histories, Claude memory files and ARTEX configuration files. The ARTEX instance ran mainly on DeepSeek v4.1-flash, likely bought through a third-party reseller, and the attacker also used GLM-5.3 from Zhipu AI and Grok 4.6 in Claude Code sessions. A Chinese-language instruction file told the model how to carry out the break-ins.

The logs show the attacker routing through nine proxy IP addresses, asking Claude where Korean breach data usually sells, and asking for help finding Korean Telegram data-sale groups. In one session the user asked Claude for a security-researcher résumé listing the results of the operation. CrowdStrike says the personal details in that prompt likely belong to the attacker but cannot yet be definitively linked, and Doom Signals is not printing them.

It is the second case in a week. SOCRadar’s CyberXero report (Oct 6) showed a Russian-speaking criminal running dozens of Claude Code agents against Ukrainian energy firms; this one shows a single operator using an open-source Chinese AI break-in tool against Korean banks. Both were found only because the attacker left AI logs exposed. CrowdStrike expects attackers to keep adding AI tooling to work faster.