Avistar logoAvistar
All insights
Security InsightsIvan Rahman8 min read

ARTEX and Machine Identities: What the FBI Found and Where Identity Discovery Fits

The FBI warns ARTEX automates intrusions at machine speed. How attackers abuse machine credentials, what teams should review, and where identity governance fits.

Share
ARTEX and Machine Identities: What the FBI Found and Where Identity Discovery Fits

Key takeaways

  • ARTEX is an AI attack framework in active use. Attackers have used it since at least August 2026 against organizations in financial services, government, technology, defense, semiconductors, energy, healthcare, manufacturing, telecommunications and legal.
  • One person sets a goal. AI agents do the rest. The operator describes an objective in plain language and the agents carry it out with minimal human input.
  • It is fast. Some intrusions went from first targeting to privileged access in under three hours.
  • It travels through suppliers. In one campaign, attackers used a compromised backup provider to attempt sign-ins to more than 2,250 downstream Microsoft 365 tenants.
  • It is not going away. The developer took the original repository down on October 8, 2026. ARTEX still spreads through forks, archived copies and a commercial version.

On November 13, 2026, the FBI published a FLASH alert on ARTEX, an open source framework that lets AI agents run an intrusion from reconnaissance through data theft 1. This article covers what the FBI and independent researchers reported, why so much of it comes down to machine credentials, and what a living identity inventory does and does not do about it.

What ARTEX is

ARTEX is a multi-agent penetration testing platform written in Go. A developer using the name "Autumn-27" released it on GitHub in July 2026 for authorized security testing, and it reached an estimated 1,500 to 2,000 GitHub stars within ten weeks 1 5.

The framework turns a high-level goal into machine commands that specialized agents execute. Google's Threat Intelligence Group describes six specialized sub-agents covering the attack lifecycle 4. ARTEX has no AI model of its own. It connects to outside model providers, and in attacker-run instances the DeepSeek Flash family was the most common choice 1 6.

Two findings from Check Point Research show how low the barrier now is. Operators mostly stayed hands-off after setting the first goal, and the estimated AI fees for three separate campaigns ran from roughly $145 to $1,500 each, based on published API prices 6.

The FBI names the practical effects 1:

  • Less skill needed. Actors with minimal expertise can run complex intrusions.
  • More targets at once. Small groups can run concurrent campaigns at a scale they could not reach before.
  • Less time to respond. ARTEX is noisy and may trigger alerts, but the attack can finish before a coordinated response starts.

How widespread it is

The FBI cites more than 1,600 publicly exposed ARTEX consoles as of early November 2026, and says the true number is likely higher once private deployments are included 1. Check Point found 2,557 systems that host or have hosted an ARTEX console, 1,608 of them reachable at the time 6. Censys, counting separately, found 1,465 hosts on November 4, about 815 of which look like distinct deployments 5.

What has happened so far

CampaignWhenWhat was reportedSources
South Korean financial institutionsLate September to early October 2026Agents exploited SQL injection and authentication bypass flaws in peripheral web applications outside core banking systems. Personal data on about 67,000 people was taken.1 7 8
ConnectWise Cloud Backup supply chainOctober 2026Microsoft attributes this to Storm-0940, a China-state-aligned actor. Compromised application credentials were used to authenticate to customer tenants through apps that already had access. Sign-ins were attempted against more than 2,250 tenants.1 2
Financial sector targeting by Storm-3358From October 2026A financially motivated, Cambodia-based group paired ARTEX with conventional tools to exploit internet-facing systems and install open-source remote access tools.3
Opportunistic cybercrimeReported November 2026Campaigns against code-signing certificate shops, crypto payment sites and gaming accounts. Entry points included hardcoded administrator passwords, plaintext credential files and leaked cloud storage credentials.6

CERT-EU, the cybersecurity service for the EU institutions, issued its own warning and noted that ARTEX exploits misconfigurations as well as traditional software vulnerabilities 10.

Why this is an identity problem

Not every ARTEX intrusion starts with a credential. The South Korean campaign began with flaws in web applications 8. But once agents are inside, and in the supply chain case from the very first step, the FBI's technique table is largely a list of things done with machine credentials 1. Three questions drive it.

Permissions: what can this credential reach?

The FBI observed ARTEX automatically mapping the effective permissions of Azure credentials, including SAS tokens, service principals, Kudu deployment credentials and storage account keys. It did this with sub-second capability checks, test writes and deletes, and deployment dry-runs. On AWS, it tested the permissions of compromised IAM users with DryRun flags 1.

The FBI describes sub-second capability checks, not a guarantee that a complete permissions map is built in under a second. The defensive question is the same: do your teams know what each machine credential can reach?

Pathways: where does that access lead?

  • Through trust. ARTEX abused cross-account cloud trust relationships from an external partner tenant and used compromised vendor credentials to pivot through trusted B2B federation 1. In the ConnectWise campaign, the applications already had access to customer tenants, so no interactive user sign-in was needed 2. Microsoft describes rapid app-only sign-in activity using compromised application credentials 2.
  • Up to privilege. Agents launched cloud compute instances with high-value IAM roles, queried cloud metadata services and forged an administrator token from unrotated authentication keys 1. Microsoft saw one management application assign itself the Global Administrator role, followed by bulk reads of secrets across Azure Key Vaults 2.
  • To persistence. Agents added credentials to existing OAuth applications and created new SSH key pairs so the access would outlast the first break-in 1.

Supply: where do the credentials come from?

The FBI lists hardcoded administrator passwords in JavaScript bundles, leaked credentials and stolen VPN configurations 1. Check Point saw plaintext credential files left in web roots and an unauthenticated API that leaked cloud storage keys 6. Microsoft reports credentials taken from public sources and compromised systems, and compromised cloud identities used for persistence and data theft 3.

Why the gap exists

Machine identities, including AI agents, outnumber human identities 109 to 1, according to Palo Alto Networks' 2026 survey of more than 2,900 security decision-makers 11. Access reviews are usually built around people. Service principals, keys, tokens and app permissions often have no named owner, and nobody is asked each quarter whether they still need what they can reach.

What the FBI recommends on identity

The alert's identity guidance includes 1:

  • Apply least privilege to multi-tenant application permissions and consent grants.
  • Monitor credential additions, privilege assignments and unusual app-only activity, especially for applications with highly privileged permissions.
  • Review audit records for credentials added to applications and service principals, new applications and owners, app-role assignments, consent grants and privileged role assignments.
  • Revoke unnecessary permissions and rotate credentials when suspicious activity is confirmed.
  • Review application configurations, environment variables, infrastructure-as-code, deployment scripts and CI/CD pipelines for embedded secrets, and rotate exposed credentials promptly.

The full alert also covers reducing the external attack surface, application protections and evidence preservation. Read it in full 1.

Each identity recommendation assumes you already know which machine identities you have and what each one can do. The FBI also recommends an inventory of internet-facing applications, including peripheral systems and supplier portals. Apply that same discipline to the identities behind them 1.

Where Avistar fits

Avistar supports ongoing machine identity governance across AWS, Azure and Google Cloud. Its living inventory gives teams visibility into service accounts, service principals, keys, tokens, app permissions and AI agent credentials, with ownership, permission context and risk prioritization. The goal is to understand dangerous access before an attacker uses it.

What ARTEX doesThe attacker's questionWhat an identity inventory gives you first
Uses sub-second capability checks to test the permissions of tokens, keys and service principals 1What can this credential reach?An inventory of discovered machine and agent identities with permission context to guide access reviews
Moves through vendor credentials, cross-tenant trust and apps that already hold access 1 2Whose trust can I borrow?Visibility into identities and permissions in your cloud tenant, so teams can review excessive access
Climbs to privileged roles through over-permissioned identities 1 2How do I get to admin?Risk prioritization to help teams review highly privileged identities first
Adds its own credentials and keys to stay 1How do I keep this access?A recorded baseline of identities and their access to compare against when you review for new credentials and role assignments

What you get:

  1. See. A living inventory of discovered machine and agent identities with permission context.
  2. Rank. Ownership context and risk prioritization, so teams know where to focus access reviews.
  3. Fix and prove. A fix plan in plain steps, with evidence for your auditors.

Findings provide evidence for access reviews and remediation work.

For MSPs and MSSPs

The ConnectWise campaign showed how one supplier's application credentials can put thousands of customer tenants in reach 2. Service providers sit on exactly that kind of trust. Avistar is built for the MSP and MSSP channel, so providers can deliver an identity inventory for each client cloud tenant and show clients which access to tighten.

What discovery does not do

Discovery is one part of the response, not all of it.

  • It does not patch the web application flaws that opened the door in the South Korean campaign 8.
  • Avistar is read only and agentless. It does not detect or stop an intrusion in progress, rotate credentials automatically or enforce runtime containment. You still need detection, response and authorized remediation.
  • It does not replace dedicated secrets scanning for code and pipelines. Permission context helps teams prioritize exposed credentials alongside those tools.

The inventory does not change access by itself. It helps authorized teams decide which permissions to remove and which credentials to rotate, reducing what an attacker could do with a compromised identity.

Five questions to answer this week

  1. Which applications and service principals hold highly privileged permissions in our cloud tenants, and who owns each one?
  2. Which vendor and multi-tenant applications have consent grants in our cloud tenants, including integrations we no longer use?
  3. Have any credentials, owners or role assignments been added to our applications and service principals since October 1, 2026?
  4. Which keys and tokens have not been rotated, and what can each one reach?
  5. Which machine identities have a path to an administrator role?

If any of these takes more than a day to answer, start with the inventory.

Next step

Book a demo to discuss your machine identity inventory, ownership gaps and dangerous cloud access.

Source note: FBI FLASH-20261113-001 is marked TLP:CLEAR and may be shared without restriction. The FBI does not endorse any company, product or service. References to the alert on this page are for information only.

Share

Frequently asked questions

What is ARTEX?
ARTEX is an open source, multi-agent penetration testing framework. The FBI reports that threat actors have used it since at least August 2026 to automate intrusions from reconnaissance through data theft with minimal human input.
Why is ARTEX a machine identity risk?
The FBI observed ARTEX testing cloud credential permissions, abusing trusted vendor access, escalating privileges and adding credentials to existing applications. A compromised machine identity can provide access without an interactive human sign-in.
What should MSPs and security teams review first?
Review highly privileged applications and service principals, vendor consent grants, credential additions, ownership changes and role assignments. The FBI recommends reviewing ConnectWise-related activity from at least October 1, 2026 and preserving evidence without delaying urgent containment.
Does Avistar stop an ARTEX intrusion?
No. Avistar is read only and agentless. It supports ongoing machine identity governance through inventory, ownership context, permission visibility and remediation guidance. Detection, response and authorized access changes remain necessary.

Sources

Bring machine identities under governance

Build a living inventory of machine identities with read only, agentless discovery.