68,000 customers' data stolen: Developer pulls the AI tool ARTEX
The open-source-based AI agent for penetration testing will no longer be updated or maintained. The decision came three days after South Korean police opened an investigation into data breaches at at least seven financial firms, in which the personal data of 68,000 people is said to have been stolen.
The developer behind the open AI tool ARTEX announced on Thursday, October 9, 2026 that the project is being shut down, citing misuse. "Out of consideration for the misuse of the tool, the ARTEX project will no longer be updated and will be converted to closed source," said the developer, who uses the GitHub handle "Autumn-27," according to Reuters. "No new versions will be released, and maintenance support will not be provided."
The same day, the developer added that ARTEX was originally built for learning and research: "It aims to help companies and organizations perform security risk testing within the scope of authorized assets and improve security protection," the statement relayed by The Hacker News reads.
The campaign that triggered the shutdown
Shortly before, in late September and early October 2026, CrowdStrike Intelligence had tracked a campaign against South Korean financial organizations in which a threat actor reportedly used ARTEX together with large language models. "In this activity, the threat actor leveraged ARTEX, a recently released open agentic pentesting tool developed in China, alongside large language models," the company wrote, as quoted by The Hacker News.
South Korea's National Police Agency opened an investigation on Tuesday, October 6 into attacks that allegedly used a Chinese-developed AI tool to penetrate at least seven South Korean financial companies and steal personal data belonging to 68,000 people, according to The Wall Street Journal, relayed via Yahoo News. Reuters has in other coverage referred to "at least nine" banks, so the exact scope remains unclear.
On some of the web servers used in the attacks, investigators found a Chinese-language text reading "ARTEX-自主渗透試控制台" — roughly "ARTEX autonomous penetration control console." On that basis, authorities are investigating whether ARTEX was used during the attacks, Seoul Economic Daily reports, citing the Journal.
What ARTEX v2 has become
Korean firm AhnLab published an analysis of ARTEX v2 on October 9 that describes a rapidly evolving tool. "ARTEX v2 is evolving from a tool that automates individual penetration tasks into a platform that coordinates and manages the work, tools and data of multiple agents in one environment," the company said, according to Seoul Economic Daily. AhnLab also found login screens on 394 systems accessible from the internet — that is, ARTEX instances exposed enough that anyone could find them.
Where it points — and how uncertain it is
CrowdStrike has not linked the activity to a named actor. "While this activity is not attributed to a named adversary, the threat actor is likely a Chinese speaker and financially motivated," the report states, as quoted by Reuters. According to Reuters, the company has expressed moderate confidence that one person behind the attacks may be a 26-year-old man in Guangdong province — an assessment that is unverified and should not be treated as an established fact.
"And this is significant because it enables one human to attack many customers in a very short time using the power of AI," said Adam Meyers, senior vice president for counter adversary operations at CrowdStrike, at a press briefing, according to Reuters.
Several details appear in conflicting light across the coverage. Shinhan Bank said, according to Reuters, that the personal data of around 25,000 customers had been compromised, while KB Kookmin Bank said data on 119 customers leaked. Yahoo News, citing the Journal, has reported the exposure of credit card data for 119,000 customers at Kookmin — a discrepancy of three orders of magnitude that coverage has yet to resolve. The developer's name is also rendered differently: "Li Puhua" at Yahoo versus "Li Fuhua" at Seoul Economic Daily, with the alias given as both "Autumn-27" and "Autumn." Information about which language models the ARTEX instances used — names such as DeepSeek v4.1-flash, GLM-5.3 and Grok 4.6 circulate in secondary write-ups of CrowdStrike's findings — cannot be independently verified from the available source material.
A first test for open AI security tools
The ARTEX case has become the first concrete governance consequence for a developer of an open AI-agent tool after criminal misuse: the tool has been withdrawn, the source code closed, and future versions will never be released. That means a tool built for "learning and research" — in the developer's own words — is now disappearing from the open ecosystem, even as the attack campaign it was linked to continues to unfold.
The open questions are many: How many banks were actually hit, and how many customers at each? Who was behind it, and what models powered the agents? And what will any eventual investigation establish? Until such details are clarified, the safest thing one can say is that an open AI pentesting tool was linked to real data breaches — and that the developer chose to close the door.