--- title: "South Korean banks: attacker relied on ARTEX AI and Claude agents" description: "According to CrowdStrike, the attacks on several South Korean banks in early October were carried out with ARTEX AI, an agentic penetration testing suite, and Claude agents. The case shows how audit tools and mainstream coding agents are now used in real intrusions." canonical: https://inyourgeek.com/en/articles/2026-10-10-banques-sud-coreennes-l-attaquant-s-appuyait-sur-artex type: article author: "Sébastien Soulier" publisher: "InYourGeek" language: en datePublished: 2026-10-10 dateModified: 2026-10-10 section: "Security" tags: ["ia", "banques", "coree", "crowdstrike"] --- # South Korean banks: attacker relied on ARTEX AI and Claude agents > According to CrowdStrike, the attacks on several South Korean banks in early October were carried out with ARTEX AI, an agentic penetration testing suite, and Claude agents. The case shows how audit tools and mainstream coding agents are now used in real intrusions. October 10, 2026 · Security · 3 min ## Several banks hit in early October On 10 October 2026, [Bleeping Computer](https://www.bleepingcomputer.com/news/security/hacker-used-artex-ai-and-claude-agents-to-target-south-korean-banks/) reported the findings of security firm CrowdStrike. According to those findings, the cyberattacks against South Korea's financial sector at the start of the month were the work of a Chinese-speaking attacker. Several banks were targeted, including Shinhan Bank, KB Kookmin Bank and Hana Bank. Customers' personal data and payment card information were exposed, and some institutions suffered service outages. The South Korean government held an emergency meeting and ordered immediate security measures for critical IT systems. For now, all of this comes from a single outlet, which is itself relaying CrowdStrike's analysis. You should read it with that caveat in mind. ## A toolchain built on several models The main tool was reportedly ARTEX AI, an agentic penetration testing suite developed in China and released as open source until recently. CrowdStrike confirms it was used. In an agentic suite, a language model drives the operations, rather than a human operator running each command by hand. The setup the researchers describe combines several models. The ARTEX instance ran mainly on DeepSeek v4.1-flash, which the attacker probably reached through an API reseller or proxy. On top of that, they used Claude Code sessions powered by GLM-5.3, from Zhipu AI, and by Grok 4.6. The Claude agents therefore served as a working environment, and the operation as a whole did not depend on a single model provider. ## The traces left on the infrastructure The attacker's own infrastructure is what allowed the operation to be reconstructed. Researchers found open directories containing Claude Code session histories, ARTEX configuration files and Claude memory files. The targets listed there match those named in earlier reports of intrusions into the financial sector. This allows CrowdStrike to link these activities with high confidence. The logs also show that the attacker had no clear plan for profiting from the stolen data. They asked Claude to suggest Telegram groups that resell data with a focus on Korea. Because they had used the same tools to write a CV, identifying details and contact information were also exposed. CrowdStrike notes inconsistencies, however, notably in the stated date of birth, and considers these details too unreliable to confirm the attacker's identity. We are not reproducing them. ## What agentic AI changes, in attack and in defence The sources describe a lone attacker, not an organised group. With a suite released for penetration testing and coding agents available to anyone, this person was able to target several major banks in a single country, while still not knowing what to do with the data obtained. The tools handle the technical side, but not the question of what to do with the data. Withdrawing the tool has its limits too. After finding that ARTEX had been used in real attacks, its developer closed the source code and stopped releasing updates. But the existing version has already spawned English and Korean derivatives, and it remains available as it stands. Finally, automation has a downside that defenders were able to exploit: an agent keeps a record of what it does. In this case, the agent's history and memory, left behind in an open directory, are what allowed investigators to reconstruct the attack step by step. ## Sources - [ARTEX AI, Claude agents used in cyberattacks on South Korean banks](https://www.bleepingcomputer.com/news/security/hacker-used-artex-ai-and-claude-agents-to-target-south-korean-banks/) — bleepingcomputer.com _Article written with AI assistance from the cited sources, reviewed and approved by a human before publication._ --- Equivalent HTML page : https://inyourgeek.com/en/articles/2026-10-10-banques-sud-coreennes-l-attaquant-s-appuyait-sur-artex Published by InYourGeek — https://inyourgeek.com/en/a-propos Method and safeguards : https://inyourgeek.com/en/methode All articles : https://inyourgeek.com/en/archives