November 19, 2026

Why We’re Using ARTEX in Every Penetration Test

ARTEX is a penetration testing framework that coordinates AI agents. Framework Security’s review of real attacks explains why we now use it in every engagement.

Why We’re Using ARTEX in Every Penetration Test

Published by

Framework Security

What is ARTEX?

ARTEX is a penetration testing framework that lets several AI agents work together on a security objective set by a person. The agents can work in parallel and share what they learn. A finding by one agent can change what the others examine next. That coordination is what we mean by an agent swarm.

Framework Security now uses ARTEX in every penetration test. We want to know what happens when weaknesses that different teams own turn up together. A vendor account left open is one finding. Evidence that the same account can still read customer records changes who needs to act and how soon.

What our November threat review found

For our November 3–17 Threat Activity Brief, Framework Security reviewed public reporting on ARTEX and related tools. Three questions kept coming up. What access survives when a supplier leaves? How quickly would someone notice a credential being misused? And can the security team tell an unsuccessful attempt from a successful one? We checked the cases below against the investigators’ accounts.

Access a supplier leaves behind

Barracuda traced a November intrusion to a cloud key it had issued to a vendor. The project had ended, but the key still worked. When the vendor’s development environment was compromised, the key was exposed in a configuration file in its GitHub repository. Barracuda says the intruder reached about ten cloud accounts over roughly 44 hours. It found several indicators consistent with ARTEX. For a client with dozens of integrations, this raises a practical offboarding question: who can verify that a supplier’s last working credential has actually been revoked? Barracuda’s incident advisory and investigation timeline.

The ConnectWise Cloud Backup case shows the reach of a trusted application. Microsoft observed Storm-0940 use ARTEX with compromised application identities to attempt sign-ins to more than 2,250 downstream tenants. Microsoft confirmed follow-on activity in multiple environments. The 2,250 figure measures attempted reach. Because the applications already had permission, the activity could take place without a user signing in. An investigation confined to employee sign-ins could miss it. Microsoft’s ConnectWise investigation.

How much time does a defender have?

In one case documented by Microsoft, five minutes elapsed between an AI agent scanning GitHub for secrets and taking data from Azure Key Vault. In its separate Storm-0940 investigation, Microsoft saw several intrusions progress from initial targeting to privileged cloud access in under three hours. Those are observed timelines in specific incidents. They give a test team a concrete question to take to the people on call: when did the first actionable signal reach someone who could revoke access? Microsoft’s exposed-secrets research.

CrowdStrike documented a different kind of speed in the VAPOR PANDA campaign. Over 11 days, one ARTEX controller directed activity against organizations across at least 14 countries. More than 200 organizations were targeted. CrowdStrike confirmed activity after compromise at nine. Recovered transcripts showed a human operator directing the agents while work proceeded against multiple targets in parallel. CrowdStrike’s VAPOR PANDA investigation.

What actually succeeded?

Two financial-sector investigations illustrate the difference between targeting and known harm. Microsoft observed Storm-3358 using ARTEX against financial institutions beginning in October. Separately, CrowdStrike documented an unidentified actor using ARTEX against South Korean financial organizations. The FBI’s November 13 advisory reported that the South Korean attacks exposed personal information belonging to approximately 67,000 people. That figure belongs to the South Korean campaign. Microsoft’s Storm-3358 report, CrowdStrike’s South Korea investigation and the FBI advisory.

Our brief also tracked AntiFox, a related platform. Arctic Wolf linked infrastructure hosting AntiFox to activity in an AWS environment. Many requests were denied. Among the successful requests was one that read a production secret. For the owner of that secret, the useful finding is which identity could read it and why that permission existed. A count of blocked requests alone would give the team the wrong impression. Arctic Wolf Labs’ incident report.

What ARTEX adds to an engagement

We start with the business question and the boundaries you authorize. For a customer portal, the question might be whether one customer’s records are adequately separated from another’s. For a cloud review, it might be whether a former supplier still has access to sensitive information. That choice determines where ARTEX can help and what evidence our testers need.

ARTEX helps organize observations about assets, identities and permissions so the tester can see which findings may be related. A suggested connection is a question for the tester. We review the underlying evidence, document what the test established within scope and identify any remaining uncertainty. That distinction matters when someone has to decide whether a fix can wait until the next maintenance window.

What that looks like at the readout

Consider a hypothetical case. Procurement confirms that a supplier contract ended last quarter. The identity team sees an enabled vendor account. The application owner sees a group with permission to read customer exports. Each team has part of the picture. The question that matters is whether the former supplier’s account still belongs to that group.

If the evidence confirms the membership and the permission, the client has a specific decision to make. The identity team can revoke the obsolete access, the application owner can review who else can read exports, and procurement can add an access check to supplier offboarding. Whether anyone actually used that access is a separate question for the logs. We want the person reading the report to know exactly which conclusion the evidence supports.

Why defenders need ARTEX too

The incidents in our brief span espionage, financial crime and the compromise of a trusted supplier. Across them, different operators chose the same framework. For a defender, the practical question is whether the team can recognize related events while they are still unfolding. In an authorized test, we can put the evidence our testers collected beside what the client’s responders saw. If the first useful alert arrived after sensitive data was already within reach, both teams know where the response needs work.

Our assessment is that ARTEX is becoming the de facto tool for offensive agent swarms. Defenders will need to use it too.

Talk With A Cybersecurity Expert

Bring the deadline you're staring at, the framework you're being held to, and the team you have. In 30 minutes we'll tell you honestly whether we're the right fit — and if we are, what the first two weeks look like.

Book a working session
Start here

A working session.
Not a sales call.

Bring the deadline you're staring at, the framework you're being held to, and the team you have. In 30 minutes we'll tell you honestly whether we're the right fit — and if we are, what the first two weeks look like.