Google tells that attackers use ARTEX for deep access in 15 minutes

Google Threat Intelligence Group and Mandiant saw attacker groups use open source AI agent tools, made to test networks, such as ARTEX, PentAGI and Hexstrike. The groups include some that want money, and Google thinks that some have a connection to China. Google tells that some attacks went from first access to deep system access in 15 minutes or less.

How it works

The tools let AI agents do the steps of an attack with small human help. ARTEX mapped Active Directory, a system that controls user accounts. It then got more rights with two known methods, Kerberoasting and DCSync, and got Azure storage keys.

The numbers

The one number is 15 minutes, for some attacks. No count of attacks, groups or victims is given.

What is not known

The agents try, make errors and try again very fast. Google tells that the attacks make many marks in logs, because they try many steps fast. It is not known how many times defenders saw these marks in time.

Insight: A closed test tool still drives attacks

A test tool was put on GitHub on 26 July 2026, and attackers used it in the next weeks. Its author closed the code, but it has more than 12,000 copies, and Unit 42 found 609 hosts that operate it. The step of the author did not stop use, and defenders must prepare for fast attackers. The first step is to find, in minutes, the use of a login by an attacker.