CrowdStrike: Attacker used ARTEX AI agent and Claude to breach South Korean banks

A single attacker relying on an open-source AI penetration testing agent was likely behind the recent wave of intrusions at South Korean financial institutions, according to research published by CrowdStrike this week. The company said the actor used the Chinese-developed ARTEX tool together with large language models, including Anthropic's Claude, in a campaign that ran from late September to early October and exposed customer personal and card data at several banks. The case is one of the clearest public examples so far of agentic AI tooling being used in real-world attacks against a critical sector.

What happened

According to BleepingComputer, the attacks hit major lenders including Shinhan Bank, KB Kookmin Bank and Hana Bank, exposing client personal data and credit card information and causing outages in some cases. iTnews reported that at least nine South Korean banks have disclosed attacks or been named by local media as targets since late September. Shinhan Bank said about 25,000 customers were affected and KB Kookmin Bank said information on 119 customers was leaked, while Infosecurity Magazine reported that Yegaram Savings Bank disclosed a breach affecting 40,000 people. CrowdStrike said the total number of affected organizations has not been confirmed.

The South Korean government held an emergency meeting, police opened an investigation and President Lee Jae Myung called for robust response measures, the publications reported. The Financial Services Commission issued a consumer alert on October 6.

Technical details

ARTEX is an open-source agentic penetration testing tool that a Chinese security engineer using the handle "Autumn" published on GitHub earlier this year, iTnews said. It is not a language model itself but connects to external models. According to Infosecurity Magazine, the attacker mainly used ARTEX to find vulnerabilities and compromise specific services: one bank's loan progress inquiry service used by brokers was breached, and at another bank an employee mobile work-support system was compromised.

CrowdStrike reached its findings by analyzing actor-controlled open directories containing Claude Code session histories, ARTEX configuration files and Claude memory files. The company said the ARTEX instance used DeepSeek v4.1-flash as its primary model, supplemented by GLM-5.3 and Grok 4.6 in some Claude Code sessions. The Claude sessions mostly involved the criminal side of the operation, such as asking where Korean breach data is typically sold and for help finding Korean Telegram data-sales groups.

In another session the attacker asked the AI to write a security researcher résumé that included a Telegram account, age, education and a location in Maoming, Guangdong. CrowdStrike said the actor may be a 26-year-old based there, but the personal details are not reliable enough to confirm identity. A man who answered a phone number listed in the report said he knew nothing about the matter, iTnews reported.

After real-world misuse was confirmed, the developer made ARTEX closed-source and stopped updates, BleepingComputer said, although English and Korean derivatives of the existing code remain available.

Attribution

CrowdStrike has not tied the activity to a named adversary. It assesses with moderate confidence that the actor is a Chinese speaker and financially motivated, based on the use of ARTEX and Chinese-language prompts. Anthropic did not immediately respond to requests for comment, according to the reports.

What to do

The Financial Services Commission advised customers of affected companies to watch for phishing and loan scams. For defenders, the case underlines that internet-facing auxiliary services, such as broker portals and employee mobile apps, can be probed and exploited at machine speed; organizations should inventory and harden such services, monitor for automated scanning patterns and prioritize patching of exposed applications.