916 Exposed AI Penetration-Testing Consoles Found Running on U.S. Hosting
SOCRadar researchers have mapped 916 live web consoles of ARTEX, an open-source framework that uses large language models to coordinate AI agents that plan security tests, run tools and record evidence of weaknesses. Starting from 1,295 U.S.-hosted candidates drawn from internet-wide scanner data, the team merged duplicates, re-checked each one and confirmed the 916 in a snapshot taken on November 11, 2026. The United States is the largest host of ARTEX worldwide.
The software was built to help security teams test systems they own, and some of these instances belong to authorised testers. However, the same framework, exposed the same way, has appeared in documented intrusions. SOCRadar also reports that a trojanized build circulating on Docker Hub gives anyone who knows a secret marker a root shell on the console host before any login is needed. Because the code is public and easy to fork, SOCRadar treats ARTEX as a software family, covering the original, its forks and its rebrands, rather than a single threat actor.
The consoles bring together assets, testing progress, captured traffic and findings. Separate research captured one console showing 1.2 million LLM tokens used on a task and an automated fetch from a malware repository, a sign that such deployments are in active use. SOCRadar's full post covers the forks, operator fleets and hunting indicators for defenders; this summary reflects only the opening of that report.