The Information Machine
Concluded·Day 2·in the edition since 9 Oct 2026·5 sources·updated 10 Oct 2026

The ARTEX South Korean bank breach

The gist

Student Used Chinese AI Tool ARTEX to Breach 7 South Korean Banks

The case shows a freely available agentic AI tool being used to breach major financial institutions at scale, with a documented AI-assisted intrusion campaign. The attacker's exposure through their own AI session histories illustrates both the operational risks attackers face and the forensic visibility that exposed server configurations can provide.

The full picture

A student at South China University of Technology breached at least seven South Korean financial institutions using ARTEX, a Chinese open-source agentic penetration testing tool, compromising personal data on 68,000 people. The attacker was caught through their own operational failures: they left directory listing enabled on their server, exposing Claude Code session histories, Claude memory files, and ARTEX configurations. Within those sessions, the attacker had asked Claude to write on their resume that they had hacked Korean banks, giving investigators personal identifying information. They also queried Claude about where stolen South Korean data could be sold.

ARTEX was created by Chinese cybersecurity engineer Li Puhua, alias 'Autumn', and has been freely downloadable on GitHub since July. It won the 'Agent+' attack and defense challenge hosted by Baidu in September. The tool can leverage multiple large language models. In this attack, the attacker ran ARTEX on DeepSeek v4.1-Flash via an API reseller and also used GLM-5.3, Grok 4.6, and Claude Code.

CrowdStrike identified an ARTEX instance at IP address 38.244.50[.]120 port 18899, which hosted an open directory containing a Claude Code markdown configuration file. That CLAUDE.md file contained a Chinese-language pentesting prompt specifying how the LLM should conduct penetration testing activities. A threat actor-controlled Hong Kong-based IP address appeared within the discovered configuration document. The actor acquired virtual private server infrastructure at multiple IP addresses for command-and-control operations. Investigators identified ARTEX's involvement by finding its HTML tag signature in command-and-control server headers. Moon Jong-hyun, security center head at South Korean cybersecurity firm Genians, publicly disclosed the findings on LinkedIn.

How it developed
9 October 2026

CrowdStrike's October 9 blog tied ARTEX, a Chinese open-source pentesting tool by Li Puhua ('Autumn') on GitHub since July, to a breach of at least seven South Korean financial institutions, 68,000 people's data compromised.

Investigators traced ARTEX by its HTML tag signature in C2 server headers and found an exposed CLAUDE.md with a Chinese-language pentesting prompt. The attacker, believed to be a South China University of Technology student, was caught after leaving directory listing enabled, exposing Claude Code sessions where they asked the AI to add the hack to their résumé.

First in the edition · 9 October 2026
8 October 2026

CrowdStrike's October 9 blog on the breach of seven South Korean banks named IP address 38.244.50[.]120 port 18899 as the ARTEX instance's host and identified a Hong Kong-based IP within the exposed configuration.

Separately, reporting identified ARTEX's creator as Li Puhua, alias 'Autumn', whose tool has been on GitHub since July and won a Baidu competition in September. Moon Jong-hyun, security center head at Genians, made the first public disclosure on LinkedIn.

First citedTransformer News
Sources
The daily email

Want this in your inbox?

I send one email each morning with the stories that moved. If you would rather just read here, that works too.

Subscribe free