---
title: AI Penetration Testing Tool ARTEX Exploited by State‑Aligned Attackers
url: https://www.elseif.net/ai-penetration-testing-tool-artex-exploited-by-statealigned-attackers
published: 2026-10-25T16:05:13+00:00
language: en
section: Models
source: https://www.itmedia.co.jp/aiplus/article/2610/23/2000002552/
organizations: ARTEX, Microsoft, AhnLab
publisher: elseif
---

# AI Penetration Testing Tool ARTEX Exploited by State‑Aligned Attackers

A Chinese‑origin open source autonomous penetration testing tool named ARTEX has been observed being used by threat actors linked to the Chinese government. Microsoft announced on October 21 that attackers employing ARTEX can automate intrusions without human intervention. South Korea and Japan have also reported confirmed misuse of the tool. The software selects appropriate AI agents from a toolset based on target, objective and permitted scope, allowing the user to choose the underlying large language model. AhnLab security research identified three cooperating agents, Goals, Planner and Worker, that operate sequentially to execute commands such as file uploads on web servers.

Tests in an isolated environment demonstrated that ARTEX can exploit public service vulnerabilities and configuration errors to gain initial access, run commands and move laterally inside networks. AhnLab confirmed that the tool can reach internal networks from a single entry point. The author of ARTEX stated in an October 8 statement that the project was created solely for academic research and was intended to assist authorized security testing within defined limits. The author denied any involvement in the observed attacks and halted further development and public releases. After the original GitHub repository was removed, modified versions of the code have begun circulating, indicating that ARTEX is being reused by multiple operators.

AhnLab identified roughly 600 public IP addresses associated with ARTEX activity, with the majority located in the United States, China, Hong Kong and Singapore. Some hosts showed traces of attack activity. The company recommends monitoring public services, limiting administrative privileges, minimizing internal network access and employing endpoint detection and response solutions to detect lateral movement. Basic defensive principles remain unchanged even when attackers leverage AI.