CrowdStrike: attacker used open-source AI pentest agent ARTEX with DeepSeek and Claude Code to breach South Korean banks

Why it matters

【Key points】

CrowdStrike reported on October 7 (US time) that an unnamed attacker used ARTEX, an open-source agentic pentesting tool recently released by a Chinese developer, plus LLMs to breach several South Korean financial institutions and take data between late September and early October.

The evidence came from open directories on the attacker's own servers: Claude Code session histories, ARTEX configs, Claude memory files and a Chinese-language pentesting prompt. ARTEX ran mainly on DeepSeek V4.1 Flash, likely via an API reseller, with GLM-5.3 and Grok 4.6 also used in Claude Code sessions. CrowdStrike assesses with moderate confidence that the actor is a financially motivated Chinese speaker, not a named group.

【Verifiable】

Primary source: CrowdStrike's blog report (October 7, 2026), with IOCs and a MITRE ATT&CK mapping. It says the number of affected organizations is unconfirmed. The WSJ reported Seoul officials saying at least seven firms were hit and data on 68,000 people stolen; Yonhap reported probes at banks including Hana, KB Kookmin and Shinhan. Those figures are from officials and media, not CrowdStrike.

【Our read】

The attacker was careless enough to leave session logs public, yet got into several banks in about two weeks. Open-source agent frameworks plus cheap API models are moving serial intrusions from team work to one-person work.

The security boundary has moved too. When most of the work runs on third-party models through a reseller inside a general coding agent, any one vendor's refusals and identity checks cover only a small part of the chain. Realistic defenses now sit with reseller identity and abuse controls, agent frameworks' defaults, and defenders running similar agents against their own systems. The agent's own logs also let investigators reconstruct the attack; they cut both ways.