ARTEX AI and Claude Agents Used in Targeted Bank Attacks
In early October 2026, a Chinese-speaking hacker launched a series of cyberattacks against several major South Korean banks, utilizing the ARTEX AI penetration testing suite and Claude agents as part of the intrusion toolkit. The attacks targeted Shinhan Bank, KB Kookmin Bank, and Hana Bank, resulting in the exposure of clients' personal data and credit card information. In some instances, the breaches led to system outages, disrupting banking services and prompting urgent concern across the financial sector.
Government and Security Response to the Breaches
The scale and sophistication of the attacks prompted an emergency meeting by the South Korean government, which called for immediate security enhancements to protect critical IT infrastructure. Authorities urged financial institutions to strengthen defenses against AI-powered threats, particularly those leveraging open-source agentic tools that have been repurposed for malicious use.
CrowdStrike Confirms Toolchain and Attacker Infrastructure
Security firm CrowdStrike confirmed that the threat actor used ARTEX AI, which until recently was an open-source agentic penetration testing suite developed in China. Through forensic analysis of the attacker’s infrastructure, researchers identified open directories containing Claude Code session histories, ARTEX configuration files, and Claude memory files. These artifacts provided critical insight into the tools and methods employed during the campaign.
LLM Backend and Proxy Usage Revealed
According to CrowdStrike, the ARTEX instance deployed in the attacks used DeepSeek v4.1-flash as its primary large language model (LLM) backend. The threat actor supplemented this with GLM-5.3 from Zhipu AI and Grok 4.6 to support additional Claude Code sessions. Researchers noted that access to DeepSeek was likely facilitated through an LLM API proxy or reseller operating at the domain xcai[.]pro.
Attacker’s Personal Data Exposed Through Resume Generation
Beyond the banking targets, the threat actor inadvertently exposed personal details by using the same AI tools to generate a résumé. This résumé included identification information, contact details, and a Telegram account handle. Based on this document, CrowdStrike assessed that the attacker may be a 26-year-old Chinese national who studied at the South China University of Technology and resides in Maoming, Guangdong. However, researchers noted inconsistencies, including an initially provided date of birth suggesting the actor was born in 2007, which casts doubt on the reliability of the identity claims.
Lack of Monetization Intent and Derivative Tools Remain Active
The investigation revealed that the threat actor did not appear to have a clear plan to monetize the stolen data. Instead, they queried Claude for recommendations on Telegram-based data sales groups focused on the Korean market. Despite the lack of immediate financial exploitation, the breach raised serious concerns about data privacy and institutional resilience.
Following confirmation that ARTEX had been used in real-world attacks, the project’s original developer decided to close-source the suite and halt further public updates. Nevertheless, existing code has been forked to create English- and Korean-language derivatives, meaning the tool remains accessible and potentially reusable by other actors.
Implications for Defending Against AI-Powered Threats
The incident underscores the growing risk posed by the misuse of advanced AI agent frameworks in cyberattacks. As tools like ARTEX AI and Claude agents become more capable and accessible, defenders must adapt their strategies to detect and respond to machine-speed intrusions. Organizations are urged to validate their security controls, monitor for anomalous AI tool usage, and prepare for rapid incident response in an era where attacks can be planned, executed, and adapted at unprecedented velocity.
Source: BleepingComputer