ARTEX: attackers paired an agentic pentest tool with Claude Code

01What happened

CrowdStrike Intelligence published this analysis on October 7, 2026, after linking infrastructure to breaches at South Korean financial organizations that began in late September 2026. Open directories on the attacker's own servers exposed Claude Code session histories, ARTEX configuration files and Claude memory files. CrowdStrike reports that data was exfiltrated.

02How it works

ARTEX is described by CrowdStrike as a recently released open-source agentic penetration testing tool developed in China. The operator ran it with DeepSeek v4.1-flash as its primary model and kept a Chinese-language CLAUDE.md file on the ARTEX host that specified how the model should conduct a pentest. Additional Claude Code sessions used GLM-5.3 from Zhipu AI and Grok 4.6. The sessions show a two-server setup, with a Hong Kong based address as primary infrastructure and the ARTEX host likely responsible for the Korean activity.

03Who is affected

Several South Korean financial organizations were breached from late September 2026, and CrowdStrike says the number affected remains unconfirmed. At one bank the operator breached a loan progress inquiry service used by financial brokers, and at another it compromised an employee mobile work-support system.

04What to do

Block or monitor the ARTEX host and the nine proxy addresses in CrowdStrike's indicator table, listed below. CrowdStrike maps the activity to three MITRE ATT&CK techniques, among them T1588.007, obtaining AI capabilities, which covers adversary use of public generative AI services.

Questions people ask

Which AI models did the ARTEX operator use?

CrowdStrike reports that the ARTEX instance ran DeepSeek v4.1-flash as its primary model, likely reached through the API proxy and reseller xcai[.]pro. The same operator used GLM-5.3 from Zhipu AI and Grok 4.6 in additional Claude Code sessions, and kept a Chinese-language CLAUDE.md file on the ARTEX host that set out how the model should run a pentest.

Did CrowdStrike attribute the South Korean finance campaign to a named group?

No named adversary is attached to this activity. CrowdStrike assesses with moderate confidence that the operator is a Chinese speaker and financially motivated, basing that on the Chinese-developed ARTEX tool and the Chinese-language prompts found in the exposed sessions. The report also notes that the operator asked Claude where threat actors usually sell Korean breach data.

What indicators did CrowdStrike publish for the ARTEX activity?

CrowdStrike lists 38.244.50[.]120 as the threat actor controlled host that ran the ARTEX instance and the open directory, plus nine proxy addresses and the xcai[.]pro API reseller. The report maps the activity to three MITRE ATT&CK techniques: T1583.003 for the virtual private servers, T1588.007 for obtaining AI capabilities, and T1090 for proxy use.

Sources 2 links, all fetched

1. 1crowdstrike.comCrowdStrike Report
2. 2attack.mitre.orgMITRE T1588.007