Hackers Use ARTEX AI Agents to Automate Network Intrusions and Gain Privileged Access in Under 3 Hour

Blog WriterCybersecurity News - Original News Source is cybersecuritynews.com


Hackers are using ARTEX, an AI-powered penetration-testing framework, to automate intrusions and shorten the time needed to gain powerful access.

In multiple attacks linked to Storm-0940, Microsoft observed attackers moving from initial targeting to privileged access in less than three hours, according to research published by ExtraHop on November 15, 2026.

Originally built for authorized security testing, ARTEX connects large language models with agents that run commands, review results, and choose their next steps.

Its developer removed the original GitHub repository on October 8 following criminal adoption, but copies and forks remain available. The FBI warning on ARTEX describes how attackers use this automation across several intrusion stages.

Researchers from ExtraHop’s Reveal Labs identified network behaviors associated with ARTEX while testing ScopeWeaver, an English-language fork, in a controlled environment.

Their technical analysis of ARTEX concerns a dual-use attack framework, not a newly discovered malware family, and separates laboratory findings from reported real-world compromises.

Hackers Use ARTEX AI Agents

ARTEX reduces manual work by dividing an operator’s objective into smaller tasks. A planner assigns those tasks to worker agents, which can run simultaneously using existing security tools.

Workers store findings in PostgreSQL, allowing the framework to adjust its approach as new information arrives or commands fail. Operators can configure local models or cloud-based services.

That workflow has supported attacks against cloud environments. ExtraHop’s report summarizes Microsoft’s findings that Storm-0940 used compromised ConnectWise Cloud Backup application credentials to attempt authentication against more than 2,250 Microsoft 365 tenants.

The ARTEX Microsoft 365 campaign exploited existing application trust; the targeting count does not mean every tenant was breached. In some environments, attackers used Microsoft Graph to discover resources, modify application credentials, and establish persistence.

ARTEX internal workflow (Source - Extrahop)
ARTEX internal workflow (Source – Extrahop)

These actions show why privileged application access matters: attackers can operate through permissions already granted to trusted software.

The under-three-hour finding describes observed attacks, not a guaranteed capability or a result reproduced by ExtraHop’s laboratory tests.

During those tests, researchers supplied valid and invalid credentials and asked ScopeWeaver to access a Windows host and obtain a domain-controller shell.

Initial SSH and RDP attempts failed. The framework then scanned with Nmap, found Windows Remote Management services, and tried alternative remote-access tools rather than stopping after its first unsuccessful attempts.

The experiment also exposed a reliability problem. ScopeWeaver confirmed WinRM connectivity but failed to establish a remote shell, yet reported success.

In a separate simulation, it successfully used Impacket’s PsExec to change a Windows lock-screen setting. That change demonstrated remote configuration control, but researchers explicitly noted that it did not establish persistence.

Detecting ARTEX Activity Before Attackers Gain Control

ARTEX’s speed does not make its activity invisible. ExtraHop detected scanning, repeated authentication attempts, remote-service activity, PowerShell remoting, and plaintext credential transmission during testing.

Its findings complement reporting on ARTEX enterprise attack automation, showing that familiar intrusion techniques remain useful detection targets even when agents select and execute them.

ScopeWeaver attempting to gain remote access to Windows host (Source - Extrahop)
ScopeWeaver attempting to gain remote access to Windows host (Source – Extrahop)

Visibility depends on where the framework runs. Internal worker messages and model requests may be observable when ARTEX operates inside a monitored network.

When attackers run it externally, those communications may never enter the victim’s environment. Inspecting supported encrypted model traffic also requires appropriate session keys; network monitoring alone does not automatically expose request contents.

Defenders should correlate bursts of failed logins, scanning, changing access methods, and remote execution involving the same systems. ExtraHop recommends investigating related events within short windows instead of relying only on infrastructure lists.

Its research into publicly exposed ARTEX consoles also highlights an important distinction: an exposed framework does not, by itself, prove criminal activity.

Indicators of compromise (IoCs):-

Type Source-reported value Interpretation
User-agent scopeweaver Observed framework string
User-agent artex-enrich/1.0 Observed framework string
User-agent Mozilla/5.0 (spa-api-recon) Observed reconnaissance string
JSON function list_findings Model-request artifact
JSON function record_fact Model-request artifact
JSON function shell_send Model-request artifact
Timing Approximately five seconds Idle-worker keep-alive interval
Lab host 10.0.4.6 Simulated Windows target
Lab controller 10.0.3.4 Simulated domain controller

Note: IP addresses and domains are intentionally Qdefanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Give your SOC the alert context on the latest threats to stop incidents quickly -> Integrate Threat Intel from 16K SOCs

The post Hackers Use ARTEX AI Agents to Automate Network Intrusions and Gain Privileged Access in Under 3 Hour appeared first on Cyber Security News.