Microsoft: China-linked attackers automate hacks with ARTEX AI tool

ARTEX matters less as a single tool than as a template. Its architecture splits the work among a Goals agent that organizes the objective, a Planner that decides the next task, and a Worker that executes tools, with the Planner adjusting based on what the Worker reports. That design means a user supplies only a target and an objective, and the system selects and runs the appropriate tools on its own.

AhnLab's isolated test showed the practical result: a plain instruction such as 'intrude into the attack target' was enough to reach remote command execution through a web server's file-upload function, with no human in the loop. The author's October 8 statement said the tool was built for academic research and authorized security testing and that the attacks were unrelated to him, and he stopped updating and publishing the project.

The removal did not end the story. AhnLab found that modified versions based on the code were being distributed, and it described ARTEX as something multiple operators reuse and redistribute rather than a single project. That points to a defensive challenge that outlasts any one repository: the capability is now loose in the wild, and the response AhnLab recommends is to tighten exposure, minimize privileges and catch the intrusion flow as it moves inward.