ARTEX AI Attacks Target South Korean Banks Using Claude Code

A Chinese-speaking attacker used AI tools to target South Korea’s financial sector, according to CrowdStrike. The ARTEX AI attacks involved an automated penetration-testing suite alongside Claude Code sessions.

Reported targets included Shinhan Bank, KB Kookmin Bank and Hana Bank. The incidents exposed customers’ personal and credit card information, while some also caused system outages.

In response, South Korean authorities held an emergency meeting and called for immediate steps to protect critical IT systems.

Exposed Files Reveal How the Attacker Used AI

CrowdStrike researchers discovered openly accessible directories on infrastructure linked to the attacker. These contained Claude Code session histories, ARTEX configuration files and Claude memory files.

Together, the records offered a detailed view of the activity. Moreover, the targets matched banks named in earlier breach reports, supporting a high-confidence connection.

According to CrowdStrike, the ARTEX installation primarily relied on DeepSeek v4.1-flash. Meanwhile, the attacker used GLM-5.3 from Zhipu AI and Grok 4.6 in additional Claude Code sessions.

Researchers also assessed that the attacker likely accessed DeepSeek through an intermediary API service. These findings distinguish the coding agent interface from the underlying models reportedly used during the activity.

Résumé Files Offer Uncertain Identity Clues

The attacker also used the same AI tools to prepare a résumé. Consequently, the exposed records included personal details, contact information and a Telegram account.

The résumé described a 26-year-old person living in Maoming, in China’s Guangdong province, who had studied at South China University of Technology.

However, another record gave a birth year of 2007, creating a significant inconsistency.

CrowdStrike therefore stopped short of confirming the attacker’s identity. Although the details may relate to the person behind the activity, the available evidence remains unreliable.

Records Suggest No Clear Plan to Sell Stolen Data

The exposed sessions also shed light on what the attacker intended to do with the stolen information.

Rather than showing an established sales strategy, the records indicated uncertainty about monetizing the data. The attacker asked Claude to suggest Telegram groups focused on selling information related to South Korea.

That request suggests an interest in finding buyers. However, it does not establish that a sale took place.

ARTEX Developer Restricts Project After Attacks

ARTEX originated in China as an open-source penetration-testing suite built around AI agents. Following confirmation of its use in real attacks, its developer made the project closed-source and discontinued further updates.

Nevertheless, existing code remains available through English- and Korean-language derivatives.

The ARTEX AI attacks illustrate how tools designed for security testing can also support malicious activity. In this case, exposed AI session records helped researchers connect the tools to specific targets and reconstruct parts of the operation.