Attacker used the open-source ARTEX pentest agent and Claude Code sessions on other models to breach South Korean banks, CrowdStrike says
Summary
CrowdStrike reported on Oct 7, 2026 that an unnamed threat actor breached South Korean financial organizations and took data between late September and early October 2026. The actor ran ARTEX, a recently released open-source agentic penetration testing tool developed in China, and also ran Claude Code sessions. CrowdStrike found the evidence in open directories on the actor's own servers: Claude Code session histories, ARTEX configuration files and Claude memory files. ARTEX used DeepSeek v4.1-flash as its main model, which CrowdStrike says was likely reached through an LLM API proxy or reseller, and the actor ran other Claude Code sessions on GLM-5.3 (Zhipu AI) and Grok 4.6. CrowdStrike assesses with moderate confidence that the actor is likely a Chinese speaker and financially motivated. Reuters reports that at least nine South Korean banks have disclosed or been reported as targets since late September. This is misuse of AI tools by an attacker, not a flaw in Claude Code or the models.
Severity
Timeline
Several South Korean financial organizations are breached; CrowdStrike dates the campaign from late September to early October 2026.
CrowdStrike publishes its report on the ARTEX campaign.
Reuters reports the CrowdStrike findings. Anthropic and South Korean police did not respond to Reuters' requests for comment.
ARTEX's developer, GitHub user Autumn-27, says the project will no longer be updated and will become closed source 'given the misuse of the tool' (reported by Reuters on Oct 9).
How it happened
CrowdStrike found an ARTEX instance on a server it ties to the activity, with an open directory holding a Claude Code CLAUDE.md file containing a Chinese-language pentesting prompt. That file pointed to a second, Hong Kong-based server, whose open directories held Claude Code session histories, ARTEX configuration files and Claude memory files. CrowdStrike describes a two-server setup: the Hong Kong server as the main attacker infrastructure and the other server running ARTEX, likely the instance behind the Korean attacks. ARTEX is not a model itself; it connects to external LLMs (per Reuters). Here its main backend was DeepSeek v4.1-flash, likely accessed through xcai[.]pro, which CrowdStrike calls a likely LLM API proxy or reseller. Claude Code sessions also ran on GLM-5.3 and Grok 4.6; the sources do not say how those models were reached. CrowdStrike also says the actor asked Claude where Korean breach data is typically sold and for help finding Korean Telegram data-sale groups. ARTEX's GitHub page says it is meant for personal learning and local testing, not real-world testing of online systems (per Reuters). CrowdStrike lists proxy IP addresses and the attacker server as indicators in its report.
How to check your app
- If you defend a financial or other high-value network, load the indicators from CrowdStrike's report into your blocking and hunting tools.
- Expect faster, broader probing: CrowdStrike says AI tooling let one actor run multiple intrusions in a short time. Patch and monitor internet-facing services accordingly.
- If you run AI coding agents or pentest agents on servers, do not expose their working directories (.claude/, session logs, configs) over the web. In this case those files exposed the attacker's own activity.
- Run open-source offensive tools like ARTEX only against systems you are authorized to test.
Sources
- CrowdStrike: Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance (Oct 7, 2026)
- Reuters (via Investing.com): South Korean banks were likely hacked by a China-based actor with an AI agent, CrowdStrike says (Oct 8, 2026)
- Reuters (via The Hindu): Chinese developer makes ARTEX AI agent closed-source after Korean bank hack (Oct 9, 2026)
- Reuters (via CNA): CrowdStrike says China-based suspect used AI tools in South Korean bank hacks (Oct 8, 2026)