AI Pentest Framework Tied to Attacks on Hundreds of Organisations

ARTEX AI is described as an autonomous penetration testing framework that uses coordinated LLM agents to carry out reconnaissance, exploitation, privilege escalation and data theft with minimal human input.

Since late August 2026 it has been observed in real intrusions, including first confirmed abuse against South Korean banks, a Microsoft 365 supply chain operation attributed to Storm-0940, and a wider campaign by Vapor Panda affecting more than 200 organisations across at least 14 countries, with nine confirmed compromises.

CrowdStrike and other researchers note that Vapor Panda ran a single ARTEX controller against hundreds of targets over 11 days starting 17 October 2026, with post‑exploitation observed at nine victims. Microsoft links ARTEX to the Storm-0940 group abusing ConnectWise Cloud Backup application credentials to access more than 2,250 Microsoft 365 tenants, while Wiz reports related activity extending to IT and software vendors.

The activity spans South Korea, Japan and other regions, including a seven‑person ring in Seoul stealing about 8.2 million records, and attacks on Japanese firms. ARTEX’s technique leans on weak identity hygiene and exposed services, using SAML and XXE chains, credential access and cloud‑identity abuse, with activity monitored through indicators such as bursts of password spraying, rapid cloud enumeration and vendor app sign‑ins from unfamiliar IPs.

Defenders are advised to focus on hardening identities, rotating long‑lived keys, and deploying behavior‑based detections; SOCRadar highlights related threat intelligence modules for asset exposure, vulnerability prioritisation and identity monitoring.