Google Sees Attackers Using ARTEX and Other Agentic Frameworks in Intrusions

Google Threat Intelligence Group detailed in a company blog post how threat actors are using ARTEX and other agentic penetration testing frameworks in active intrusions. Since late August 2026, Google and Mandiant have observed ARTEX, PentAGI, CAIRN and possible custom tools handling reconnaissance, lateral movement, privilege escalation, database exploitation and preparations for data theft.

ARTEX splits the attack lifecycle across six specialized sub-agents and uses frontier and open source models to translate high level tactical goals into machine commands. In one operation, it produced 27 events in 23 seconds of active burst time while probing AWS permissions. In another, it adapted an oversized script and deployed a 17 snapshot instance in about 10 seconds. Its trial and error approach can also create extensive logs, including 240 events during attempts to gain command execution through a web server.

The framework has been used by financially motivated extortion groups and suspected China linked operators against organizations in the United States, Europe and Asia. A campaign Google tracks as UNC7228, which overlaps with the activity Microsoft calls STORM-0940 and which Wiz has also tracked, typically reached deep system access in about four hours after the first account compromise, and in one case within 15 minutes. The original ARTEX repository was removed from GitHub in October 2026, but community forks and commercial variants remain available.