DURING October 2026, Wiz Research tracked an actor leveraging the ARTEX AI agent to probe IT and software vendors, aligning with Microsoft’s Storm-0940 reporting that Chinese state-aligned groups had been involved in a supply-chain compromise using stolen vendor credentials. ARTEX is an open-source autonomous penetration-testing tool released on GitHub in July 2026; a planner agent selects steps and worker agents run tools such as nmap and Impacket. The activity quickly spread to attacks on multiple vendors, with apparent targets including internal identity providers.
Wiz’s observations show a repeatable playbook: the agent downloads JavaScript bundles, source maps and API docs, then targets internal routes revealed by those maps. It enhances password-recovery testing, registers throwaway emails on sign-up where possible, and then attempts to view other users’ data. It also tested forged, unsigned login tokens against more than 60 SSO routes in two minutes, and conducted password spraying bursts against Okta and Entra ID with new user agents and one-second pacing.
The attacker aimed at software engineers’ accounts, and sessions varied in length from minutes to 24–30 hours; when a defender blocked an IP, a new one appeared within minutes.
The incident underscores the risk to vendor ecosystems: even if most attempts failed, continued iteration by an autonomous agent could expose downstream customers. Recommendations from Wiz and related reporting include restricting public source maps, rate-limiting sensitive endpoints, rejecting unsigned tokens, blocking disposable emails, and validating tenant isolation.