AI-powered tool from China used to breach multiple South Korean banks, CrowdStrike finds
An unidentified hacker, likely a Chinese speaker, used ARTEX — an open-source AI penetration-testing tool — to hit major South Korean banks in days, exposing data on tens of thousands of people.
An unidentified hacker tore through multiple South Korean financial institutions in a matter of days last week, using an AI-powered penetration-testing tool built in China to automate attacks that would have taken human hackers weeks to plan and execute — exposing the personal data of tens of thousands of people and prompting police and financial regulators to open investigations.
U.S. cybersecurity firm CrowdStrike, in a report released Wednesday, identified the tool as ARTEX — an open-source, AI-driven penetration-testing platform developed in China and distributed primarily through GitHub. The attacker used ARTEX alongside large language models (LLMs) to carry out the attacks between late September and early October, according to CrowdStrike.
While this activity has not been attributed to a named adversary, the threat actor is likely a Chinese speaker and financially motivated.— CrowdStrike
CrowdStrike said that assessment was made with moderate confidence, based on the use of ARTEX and observed Chinese-language prompts found in the attacker's sessions. Two servers were tied to the attacks: one in Hong Kong that served as the main infrastructure, and a second on which ARTEX was hosted.
The confirmed victims include Shinhan Bank, KB Kookmin Bank, Hana Bank, BNK Busan Bank, and Yegaram Savings Bank. According to the Korea Financial Security Institute, Woori Bank and NH NongHyup Bank each detected intrusion attempts but avoided a breach, as their systems lacked the particular vulnerabilities the attacker was targeting.
None of the attacks touched customer-facing banking services; all targeted internal systems used by employees or partners. At Shinhan Bank, a loan-agent inquiry service was the entry point, with 25,729 individuals' data exposed. An employee mobile work-support system at KB Kookmin Bank yielded 119 leaked records. Hana Bank's employee sales-support system was the source of 89 records taken. At BNK Busan Bank, details belonging to 11 contract workers were exposed. The confirmed total currently stands at just under 26,000 affected individuals, but the Korea Financial Security Institute says that figure could rise as investigations continue — two additional savings banks are under active review.
The gap between customer-facing and internal systems is at the heart of why the attacks succeeded where they did. South Korean banks have heavily fortified their public-facing digital services in recent years, but internal employee tools operated under looser security standards — and across a wide enough scope that the banks themselves had difficulty identifying where vulnerabilities existed, according to the Korea Financial Security Institute.
Security on customer-facing electronic financial services has been enormously strengthened, but internal employee-facing systems had been managed less rigorously. There are so many management points across such a wide scope that financial institutions themselves appear to have had difficulty identifying where those vulnerabilities existed.— Korea Financial Security Institute official
What made the attacks particularly striking to security experts was their speed and breadth. Lim Jong-in, a professor at Korea University's Graduate School of Information Security, said that for human hackers alone, moving so quickly across the financial sector within just a few days would have been nearly impossible.
It would take weeks, if not months, for human hackers to plan and carry out such large-scale attacks.— Lim Jong-in, professor, Korea University Graduate School of Information Security
ARTEX is designed to automate tasks that typically require human expertise — scanning systems for vulnerabilities, developing potential attack paths, and probing networks. The tool supports connections to AI models from DeepSeek, a Chinese company, along with those from OpenAI and Anthropic, enabling several AI agents to assess targets at the same time. According to CrowdStrike, the attacker relied chiefly on DeepSeek v4.1-flash, with GLM-5.3 and Grok 4.6 as supplements, operating through Claude Code sessions developed by Anthropic.
During one Claude Code session, the attacker directed the AI to produce a security researcher resume drawing on personal details that included an age of 26 and studies at South China University of Technology in Guangdong, China. CrowdStrike's analysis of files further revealed that the attacker used Claude to search for marketplaces dealing in stolen South Korean data and Telegram groups engaged in selling such information, pointing to a financial motive.
After examining attack IPs and server logs from Shinhan Bank — the first institution to report an incident — the Korea Financial Security Institute was able to confirm ARTEX's role in the campaign. Once those IPs were shared with the broader financial sector, other banks reviewed their own access logs and discovered breaches they had not previously detected. The attacker has continued operating by rotating IPs when one is blocked.
Blocking IPs is little more than emergency first aid. When you analyze the attack data, there is a specific data signature common to traffic originating from ARTEX, which allows us to identify the attacker.— Korea Financial Security Institute official
The institute stressed that despite the data exposure, the risk of direct financial loss to consumers is considered low. The attacker did not take over systems — they queried and extracted information. Officials said the more pressing worry is that the stolen personal data could fuel voice phishing scams, and that comparable attacks could be launched again at any time using any of a range of similar open-source AI tools.
There are a great many open-source AI tools like ARTEX. There is currently no way to restrict everything being developed and distributed around the world.— Korea Financial Security Institute official
Hwang Suk-jin, a professor at Dongguk University's Graduate School of International Affairs and Information Security, cautioned that the relatively contained damage seen so far may not represent the full picture. He raised the possibility that the recent attacks served as a preliminary probe to map out weaknesses, which could lay the groundwork for larger follow-up attacks drawing on the data already gathered.
Banks have core networks and supporting networks. The supporting networks have relatively weaker security, and they were targeted in this attack. Rather than entering through the front door, the attackers came in through a window, using an indirect route to steal only some of the information.— Hwang Suk-jin, professor, Dongguk University Graduate School of International Affairs and Information Security
Lim Jong-in cautioned against concluding that China as a state was behind the attacks simply because a Chinese-language tool was used. He pointed out that certain Chinese AI models are made available with comparatively limited safeguards, meaning anyone could download and adapt them for harmful use.
South Korean financial regulators are now preparing follow-up measures. A recommendation currently being drafted by the Korea Financial Security Institute would require financial firms to carry out a thorough audit of all internal employee-facing systems that have external access points. The Financial Services Commission is expected to be involved in those follow-up steps. Given that the attacks originated abroad, the institute has worked alongside police from the time the first report came in.
Why it matters — This is among the first documented cases of an open-source AI tool being used to automate large-scale intrusions into the financial system, signaling that the barrier to sophisticated cyberattacks has dropped significantly — and that internal bank systems long considered lower-priority targets are now squarely in the crosshairs.
⚠ Not yet confirmed
- The attacker may be a 26-year-old based in Guangdong province, China, based on personal details used in a Claude Code session to draft a fake security researcher resume
- Total affected individuals may be 60,000–68,000
- ARTEX was developed by a Chinese engineer known by the alias Autumn (Li Puhua)
- The tool's developer added anti-misuse guidelines after the incident
- South Korean President Lee Jae-myung urged immediate cybersecurity steps in response
- at least seven (changed to multiple)
Sources differ on Whether China as a state or Chinese nationals are responsible for the attacks: CrowdStrike assesses with moderate confidence the attacker is likely a Chinese speaker, based on tool use and language prompts (en.yna.co.kr) vs Korea University professor Lim Jong-in cautions that Chinese AI models have weak safeguards and could be used by anyone worldwide; the actor remains unknown (koreatimes.co.kr)
Reported by archive.is, qz.com, en.yna.co.kr, reuters.com, biz.heraldcorp.com, koreatimes.co.kr, asia.nikkei.com
- Lee Jae MyungPresident of South Korea since 2025