ARTEX: Chinese AI Hacking Agent Behind Korea Bank Leaks Hits Japan
3 min readThe same Chinese-built AI agent that Korean regulators tied to data leaks at 16 financial institutions last month has now been traced to attacks on roughly 80 Japanese companies. Tokyo security firm GMO Cybersecurity by Ierae said on Monday it found the ARTEX AI hacking tool on all six overseas servers used in a month-long campaign against Japanese insurers, financial firms and logistics companies.
Background: a pen-test tool released as open source
ARTEX is an autonomous penetration-testing program written by Chinese engineer Li Fuhua, who built it as a side project while working as a threat detection engineer at security company Threatbook. Rather than shipping its own model, it chains commercial large language models, including Anthropic’s Claude, OpenAI’s GPT and China’s DeepSeek, so an agent can analyze a target, map a route in and run security tools on its own. Li published the code on GitHub in July. It was meant for companies testing their own defenses, but anyone could download it, and someone did.
What the ARTEX AI hacking tool did
CrowdStrike found that attackers paired ARTEX with large language models in intrusions at South Korean financial institutions between late September and early October. Korean regulators say about 100,000 records leaked from 16 institutions, including Shinhan Bank, KB Kookmin Bank, Hana Bank and Busan Bank, and the National Police Agency’s cyber bureau is investigating. The developer confirmed the misuse, took the repository private on October 8 and stopped updates.
By then the campaign had already moved. According to Seoul Economic Daily and Bloomberg, GMO’s investigators, working at the request of affected companies, identified six public servers that hit Japanese businesses between September 18 and October 18. Each carried ARTEX traces and no sign of any other agent. Targets spanned e-commerce, finance, retail, logistics, media, education and marketing, and eight companies have already disclosed leaks. Japan’s incident response body JPCERT/CC says it has seen AI agents in attacks on Japanese firms since August. Separately, US firm Noforge reported that another Chinese open-source agent framework, Cairn, was used against online retailers in Japan, the US and Australia, with AI usage costs averaging about $5 per attack.
Why it matters
Daiki Fukumori, a former Interpol investigator who led GMO’s probe, said the victims were attacked through minor system vulnerabilities, the kind that used to be too tedious to exploit at scale. An agent that grinds away for 23 hours at a few dollars of model cost changes that math, and the campaigns appear to be organized by country and by industry.
The tool is not going away either. Threatbook CEO Xue Feng has announced a commercial edition, ARTEX Pro, on a cloud platform called RedAgent.ai, with a waitlist already open. Watch for Japanese regulators to follow Korea’s lead, and for model providers to face harder questions about who is calling their APIs and why.