AI hacking tool ARTEX hits South Korean banks, turns up in Japan too
The open-source tool ARTEX lets external AI models hunt for vulnerabilities automatically. It was abused in attacks on South Korean financial institutions, and GMO Cybersecurity also found it in attack infrastructure in Japan. The case shows how AI makes attacks cheaper and faster, without any new security flaws.

What this is about
ARTEX was used in attacks on South Korean banks and other financial institutions. It is an open-source tool for automated penetration testing (authorized attack simulations that look for vulnerabilities). CrowdStrike and Zscaler analyzed the attacks. On October 19, GMO Cybersecurity by Ierae reported that it had also found the tool at an attack source that targeted Japanese companies.
How ARTEX works
ARTEX has no AI model of its own. It drives existing large language models (ChatGPT, Claude, DeepSeek) as a team of agents, according to the Wall Street Journal as cited by the Herald. It was developed by Li Puhua (alias Autumn-27) from China. According to ThreatBook, it was a personal project of the product lead of its RedAgents platform. It won an attack-and-defense competition run by Baidu (among about 150 teams, according to Korean media).
The architecture consists of "Goals" (break objectives down), "Planner" (decide the next step), "Worker" (run tools, up to 3 in parallel by default) and "mainagent" (human intervention). Knowledge is stored in an asset map (systems and access) and an exploration map (facts and dependencies). Every update wakes the Planner again.
The attack in South Korea
According to CrowdStrike, the campaign ran from late September to early October against at least nine institutions. The main model was DeepSeek v4.1-flash, supplemented by GLM-5.3 and Grok 4.6. The attacker also used Claude Code. Because he left open directories on a server, investigators found histories, configurations and questions about where to sell Korean data and how to find suitable Telegram groups. There were two servers (Hong Kong plus one for ARTEX). CrowdStrike assesses with medium confidence that the attacker is Chinese-speaking and financially motivated, with no attribution to a specific group. Reuters, as cited in a Yahoo report, names a 26-year-old from Guangdong, who denies everything.
The steps: search for peripheral systems (query portals, mobile back ends for employees), break in via SQL injection (smuggling database commands into inputs) or authentication bypass, exfiltrate names, phone numbers, income and credit limits, then look for buyers. The financial security agency said the AI was not an autonomous actor; hackers used it as a tool. The attacks ran through more than 20 IP addresses in over ten countries.
Numbers and damage
| Source | Figure |
|---|---|
| CrowdStrike | at least 7 companies |
| Zscaler | approx. 67,000 people |
| FSC (Oct. 14) | 16 companies, approx. 100,000 records |
At Shinhan Bank, 25,729 people were affected; KB Kookmin 119, Hana 89, BNK Busan 11. According to the FSC, passwords and card numbers were not affected. The entry points were systems for employees and partners, not customer-facing online banking.
On October 29, CrowdStrike also reported that the China-linked group VAPOR PANDA used ARTEX and DeepSeek to attack more than 200 organizations in over 14 countries within 11 days. A connection to Korea or Japan has not been established.
Traces in Japan
GMO found an attack source using ARTEX that targeted about 20 Japanese companies, 9 of them from the same industry. Six more AI attack servers targeted retail, education, insurance, logistics and other sectors; 8 targets had already reported incidents in 2026. Media reports put the total at about 80 companies. According to the Epoch Times, GMO confirmed five more ARTEX servers. The blog author points out, however, that only the first source is explicitly attributed to ARTEX in GMO's blog. Reported data leaks (Lawson, Yamato, GMO Research & AI) have not yet been publicly linked to ARTEX.
Safeguards and countermeasures
Zscaler says ARTEX blocks government and university domains, but not financial institutions. The developer made the code closed-source on October 8, but forks are still circulating. Recommended measures: fewer public-facing services, permission checks on every request, rate limiting, monitoring of connections to AI APIs, and isolation of testing tools.
Takeaway
AI doesn't create new vulnerabilities, but it lowers the effort and skill needed to attack. The attacker's identity and the full extent of the activity in Japan remain unknown.
Related videos
Sources
- Bing News (ja-JP) note.com
- yahoo.com yahoo.com
- theepochtimes.com theepochtimes.com
- biz.heraldcorp.com biz.heraldcorp.com
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.