Home » Hackers Use ARTEX AI Agents to Automate Network Intrusions and Gain Privileged Access in Under 3 Hours
Artificial intelligence is changing the cybersecurity landscape, and attackers are increasingly using autonomous AI agents to automate activities that previously required significant manual effort. From network reconnaissance and credential testing to privilege escalation and lateral movement, AI-powered frameworks can coordinate multiple attack stages with limited human intervention. One such framework, ARTEX, is now drawing attention after researchers identified its use in real-world cyber intrusions, demonstrating how AI automation can accelerate attacks against enterprise networks and cloud environments.
As per a report published by Cybersecurity News on November 16, 2026, citing research from ExtraHop’s Reveal Labs, threat actors are using ARTEX, an AI-powered penetration-testing framework, to automate network intrusions and shorten the time required to gain privileged access. In multiple attacks associated with Storm-0940, Microsoft observed attackers progressing from initial targeting to privileged access in less than three hours. The report also explains how researchers tested an English-language fork called ScopeWeaver in a controlled environment, identifying network behaviors that security teams can monitor to detect similar automated intrusion activity.
ARTEX is an AI-powered penetration-testing and offensive security framework originally designed for authorized security assessments. It connects large language models (LLMs) with specialized agents that can execute commands, analyze results, and determine subsequent actions based on the information collected.
Unlike conventional security tools that require an operator to manually execute individual commands, ARTEX uses a coordinated agent architecture to divide an objective into smaller tasks.
Its operational workflow includes:
The framework supports both locally hosted and cloud-based AI models, allowing operators to configure the models used during different operations.
This architecture makes ARTEX particularly concerning when used by unauthorized threat actors. It can reduce the time spent on repetitive reconnaissance, tool execution, and decision-making, allowing attackers to coordinate several intrusion activities at once.
ARTEX is a dual-use security framework, not a newly discovered malware family. Its original developer removed the GitHub repository on October 8, 2026, following reports of criminal adoption. However, copies and modified forks continue to circulate.
One of the most concerning examples involves Storm-0940, a China-linked threat actor associated with attacks against Microsoft 365 environments.
According to Microsoft’s findings summarized in the ExtraHop report, attackers used compromised ConnectWise Cloud Backup application credentials to attempt authentication against more than 2,250 Microsoft 365 tenants.
The campaign highlights the danger of compromised third-party application identities. Rather than relying exclusively on conventional interactive logins, attackers abused existing application trust and permissions to access connected cloud environments.
In some environments, attackers used Microsoft Graph to discover resources, modify application credentials, and establish persistence.
The attack demonstrates how a compromised service provider or trusted application can become an entry point into customer environments.
Importantly, the 2,250 figure represents tenants targeted for authentication attempts, not 2,250 confirmed breaches. Similarly, the under-three-hour finding describes observed attack activity rather than a guaranteed capability of ARTEX in every environment.
ExtraHop’s Reveal Labs conducted controlled experiments using ScopeWeaver, an English-language fork of ARTEX, to understand how the framework behaves during network intrusion attempts.
Researchers supplied valid and invalid credentials and instructed the framework to access a Windows host and obtain a shell on a domain controller.
The experiment revealed how the AI agents adapted their approach when initial remote-access attempts failed.
The researchers initially observed the framework attempting to access the Windows environment using SSH and Remote Desktop Protocol (RDP).
These attempts failed, but the framework did not simply stop its operation. Instead, it evaluated the available network information and attempted alternative methods.
The framework used Nmap to scan the target environment and identify available services.
The scan revealed Windows Remote Management (WinRM) services, providing another potential route for remote access.
This demonstrates how an AI agent can use the output of one security tool to determine which tool or technique should be attempted next.
After identifying WinRM, ScopeWeaver attempted to use alternative remote-access tools to establish a connection.
The framework successfully confirmed WinRM connectivity but failed to establish the intended remote shell.
An important finding was that the agent incorrectly reported success despite failing to complete the remote-access operation. This exposed a reliability limitation in AI-generated execution reports.
In a separate controlled simulation, researchers observed ScopeWeaver successfully using Impacket’s PsExec to modify a Windows lock-screen setting.
This demonstrated that the framework could execute remote configuration changes when the necessary access was available.
However, researchers explicitly noted that this activity did not establish persistence. It should therefore be understood as evidence of remote configuration control rather than proof of a persistent compromise.
Observed laboratory activity:
Initial SSH/RDP attempts → Nmap scanning → WinRM discovery → Alternative remote-access attempts → PsExec execution in a separate simulation
The distinction between successful and unsuccessful actions is important. The laboratory findings demonstrate the framework’s behavior and potential capabilities, not a confirmed successful domain-controller compromise in that experiment.
ARTEX does not necessarily depend on a completely new vulnerability or an unfamiliar attack technique. Instead, it automates existing security tools and combines familiar intrusion methods into a coordinated workflow.
This creates a significant challenge for security operations teams.
An individual Nmap scan, failed login, or PowerShell command may appear legitimate when examined separately. However, when multiple activities occur rapidly against the same environment, they may indicate an automated intrusion.
ExtraHop identified several network behaviors associated with ARTEX activity during its testing.
Observed behavior | Security significance |
Network scanning | May indicate automated reconnaissance and service discovery |
Repeated authentication attempts | Can indicate credential testing or unauthorized access attempts |
SSH and RDP attempts | May reveal attempts to establish remote access |
WinRM activity | Can indicate remote Windows administration or lateral movement |
PowerShell remoting | May indicate remote command execution |
Plaintext credential transmission | Can expose authentication information and create opportunities for credential theft |
Rapid changes in access methods | May indicate an automated framework adapting after unsuccessful attempts |
The research also highlights an important limitation of network-based detection.
When ARTEX operates within a monitored network, internal worker communications and model-related requests may provide useful visibility. However, if attackers operate the framework externally, those communications may never enter the victim’s network.
Furthermore, inspecting encrypted model traffic requires appropriate session keys. Network monitoring alone does not automatically reveal the contents of encrypted AI requests.
For defenders, the priority should be correlating observable attack behaviors rather than depending exclusively on identifying the ARTEX framework itself.
ExtraHop identified several indicators associated with ScopeWeaver and ARTEX activity during its controlled analysis.
Type | Source-reported indicator | Description |
User-Agent | scopeweaver | Framework-related user-agent string |
User-Agent | artex-enrich/1.0 | Framework-related user-agent string |
User-Agent | Mozilla/5.0 (spa-api-recon) | Reconnaissance-related user-agent |
JSON function | list_findings | Model-request artifact |
JSON function | record_fact | Model-request artifact |
JSON function | shell_send | Model-request artifact |
Timing | Approximately 5 seconds | Idle-worker keep-alive interval |
Laboratory host | 10.0.4.6 | Simulated Windows target |
Laboratory controller | 10.0.3.4 | Simulated domain controller |
These indicators should be treated as investigation leads, not definitive proof of malicious activity. The two IP addresses belong to the researchers’ simulated environment and should not be interpreted as attacker infrastructure.
Security teams should combine indicator-based monitoring with behavioral analysis, authentication monitoring, and endpoint investigation.
The emergence of AI-powered offensive frameworks reinforces the importance of detecting attack sequences rather than relying exclusively on individual indicators.
Organizations should strengthen their defenses across identity, network, endpoint, and cloud environments.
Security teams should monitor repeated login failures, unusual authentication patterns, unexpected account access, and suspicious changes in authentication behavior.
Privileged accounts should receive additional monitoring because compromised administrative credentials can provide attackers with access to critical infrastructure.
Services such as SSH, RDP, and WinRM should not be unnecessarily exposed to untrusted networks.
Organizations should restrict remote administration to approved systems, enforce strong authentication, and monitor unusual remote execution activity.
Domain controllers and privileged identities are high-value targets. Organizations should monitor suspicious PowerShell activity, unusual directory enumeration, unauthorized privilege changes, and attempts to access sensitive authentication material.
The Storm-0940 campaign highlights the risk of compromised application credentials.
Organizations should regularly review OAuth permissions, service principals, application credentials, and third-party integrations. Unnecessary permissions should be removed, while credential rotation and access reviews should be performed regularly.
ARTEX can execute multiple activities rapidly. Security teams should correlate network scanning, failed authentication, remote service discovery, and command execution involving the same systems.
A series of seemingly unrelated events occurring within a short period may reveal an automated intrusion attempt.
Network segmentation and restricted access to administrative infrastructure can limit the opportunities available to attackers after initial access.
Critical systems, domain controllers, and privileged management environments should be isolated from ordinary user networks wherever practical.
The ARTEX campaign demonstrates why modern security operations require continuous visibility across endpoints, identities, networks, and cloud infrastructure. AI-powered attack frameworks can accelerate reconnaissance and automate familiar techniques, making it essential for organizations to correlate suspicious activities and identify threats before attackers gain privileged access.
Seceon’s AI-driven cybersecurity platform provides capabilities that can support this layered defense strategy.
Seceon aiSIEM can collect and correlate security events from integrated network devices, authentication systems, servers, endpoints, and cloud environments.
In an ARTEX-like scenario, this centralized visibility can help security teams investigate combinations of network scanning, repeated authentication failures, suspicious PowerShell activity, and unexpected remote-access attempts.
By correlating events across multiple sources, security teams can gain a clearer understanding of whether isolated activities represent a coordinated intrusion.
Seceon aiXDR-PMax provides endpoint-focused threat detection and investigation capabilities.
It can help identify suspicious process execution, malicious files, abnormal endpoint behavior, and activities associated with unauthorized remote execution where relevant endpoint telemetry is available.
This is particularly important when attackers move from network reconnaissance to executing commands on compromised Windows systems.
The Storm-0940 campaign highlights how compromised application credentials and excessive permissions can become entry points into cloud environments.
Seceon aiSecurity UID Guard360 can support identity and access security by helping organizations improve visibility into identity-related risks and suspicious access activity.
ARTEX-associated attacks demonstrate the importance of protecting cloud applications, service identities, and connected infrastructure.
Seceon aiSIEM-CGuard can support security visibility across integrated cloud environments, helping teams investigate suspicious activity involving cloud resources and connected services.
Exposed management interfaces, publicly accessible services, and weak external configurations can provide opportunities for automated reconnaissance.
Seceon aiSecurityScore360 can support external attack surface and security posture assessments, helping organizations identify exposed assets and prioritize weaknesses requiring remediation.
Together, these capabilities can help organizations strengthen security monitoring, investigate coordinated intrusion attempts, and improve visibility across hybrid environments. Their effectiveness depends on available telemetry and appropriate integrations, and they complement rather than replace identity hardening, network segmentation, secure cloud configurations, and endpoint protection.
The growing use of ARTEX demonstrates how AI-powered agents are changing the speed and scale of enterprise cyber intrusions. By connecting large language models with security tools, the framework can automate reconnaissance, evaluate command outputs, and adjust its approach when initial attempts fail.
The Storm-0940 campaign highlights the potential consequences of this automation, particularly when attackers abuse compromised third-party application credentials to access cloud environments. ExtraHop’s controlled experiments further demonstrate how frameworks such as ScopeWeaver can discover network services and attempt alternative remote-access methods.
For organizations, the challenge is not simply detecting a particular AI tool. It is recognizing the familiar attack techniques that autonomous agents can execute at machine speed.
Continuous monitoring, identity protection, endpoint visibility, cloud security, and behavioral correlation are essential to identifying these threats early. As AI-driven offensive frameworks become more accessible, organizations must strengthen their ability to detect coordinated activity across the entire enterprise environment before attackers turn initial access into privileged control.
Copyright @Seceon Inc 2026. All Rights Reserved.