Censys Finds ARTEX AI Hacking Framework on 1,465 Public Hosts

Censys identified 1,465 internet-facing hosts exposing ARTEX, an LLM-driven autonomous penetration-testing framework released in July 2026. Repeated hosting configurations across approximately 650 hosts suggest cloned systems or multiple addresses, leaving roughly 815 distinct-looking deployments, not 1,465 independent operators. Researchers found 32 open directories exposing prompts, configuration files, exploitation tools, stagers, implants, and operator files; 155 hosts also carried other security-tool or threat detections, including command-and-control, phishing, and tunneling tools.

Separate investigations reportedly linked ARTEX to two malicious campaigns disclosed in October: CrowdStrike connected it to attacks against South Korean financial institutions, while Microsoft attributed its use in a ConnectWise Cloud Backup supply-chain compromise affecting customer tenants to China state-aligned actor Storm-0940. ARTEX exposure and tool co-location alone do not establish malicious activity, because authorized penetration testers also use the framework. Defenders should correlate infrastructure indicators with authentication logs, endpoint activity, network connections, and historical exposure records before classifying deployments as hostile.