ARTEX AI Pentesting Tool Tied to Korean Financial Data Theft
CrowdStrike Intelligence has detailed a targeted campaign against South Korean financial organizations in which attackers used ARTEX, an open-source agentic AI penetration-testing tool, alongside large language models. The activity ran from late September to early October 2026 and resulted in data exfiltration.
The campaign has not been attributed to a known threat actor or group. CrowdStrike said the available evidence points to a suspected Chinese-speaking operator with a financial motive, while stressing that the information does not definitively identify the person behind the activity.
Exposed operational files revealed the infrastructure
CrowdStrike discovered the activity after finding open directories on a Hong Kong-based IP address. Those directories exposed Claude Code session histories, Claude memory files and ARTEX configuration files, providing a view of how the operation was assembled.
The researchers described a two-server architecture. The Hong Kong-based address served as the campaign backbone, while IP address 38.244.50[.]120 hosted the ARTEX instance suspected of being used in attacks on Korean targets.
ARTEX is a large-language-model, multi-agent autonomous penetration system developed by Autumn-27. In the observed configuration, it used DeepSeek v4.1-flash as its main LLM backend, with Z.ai's GLM-5.3 and SpaceXAI's Grok 4.6 supplementing the model. CrowdStrike suspects that DeepSeek access was obtained through the likely API reseller xcai[.]pro.
Sessions indicated interest in selling stolen data
In addition to ARTEX operations, the operator asked Claude where threat actors typically sell Korean breach information and sought assistance locating Korean Telegram groups for data sales, CrowdStrike said. One Claude Code session referenced the Telegram account @YY520CN and the name YY.
Other sessions concerning vulnerability research on a Telegram-based NFT gift marketplace used the same username. CrowdStrike said the personal details in the prompt likely belong to the operator, but that the current evidence cannot conclusively connect them to the ARTEX activity.
Developer closes the project after reported abuse
Autumn-27 said ARTEX had been intended for learning, research and authorized enterprise security risk testing. Following the reported malicious use, the developer said the project would no longer receive updates, would become closed source, and would have no future version releases or maintenance support.
The development illustrates how autonomous tooling can combine model reasoning with penetration-testing workflows in an intrusion context. For financial organizations, the practical implication is to treat access to AI-enabled testing systems as privileged: limit use to authorized assets, retain activity logs, and review exposed development or configuration files that could reveal operational detail.