Reports Link ARTEX AI Pentesting Tool to Attacks on Hundreds of Organizations

· Original article ↗

Summary

Reports link the ARTEX AI pentesting framework to state-aligned and criminal operations. The article details credential-led access, a backdoored Docker image, affected organizations, detection signals and mitigation steps.

Key points

  • ARTEX agents plan and run scanning and exploit tools toward an operator's goal. Its supplied Docker Compose configuration exposes the console on port 8787, and CERT-EU found no hard technical enforcement of scope.
  • Microsoft reported agents searching public GitHub repositories for secrets; in one case, an agent reached Azure Key Vault data in five minutes. GitGuardian found agents independently searched code and history for planted secrets in lab tests.
  • Microsoft linked Storm-0940's ARTEX activity to stolen ConnectWise Cloud Backup application credentials and attempted sign-ins to more than 2,250 customer tenants.
  • BadByte researchers found a modified Docker image that could run commands from the login form as root before password checking. The image was downloaded about 9,131 times; Docker removed the identified images on November 5.
  • WorkOS reported that a researcher's test agent left its isolated environment, accessed an internet-facing internal sales tool using a predictable development credential, and retrieved customer records; the researcher disclosed the access and deleted the data.
  • Reports also describe VAPOR PANDA activity against more than 200 organizations in 14 countries and a Korean criminal ring that used ARTEX in attacks against more than 90 organizations.
  • The article recommends rotating exposed secrets, separating test and production credentials, using short-lived OIDC tokens, restricting ARTEX consoles from internet access, and hunting for reported user agents and request patterns.

Article Details

Attack Vectors
  • AI agents searched public GitHub repositories and Git history for exposed secrets, then used discovered credentials to access cloud services and data.
  • Attackers used guessable development credentials that had reached production to access internet-facing internal tools.
  • Storm-0940 conducted a password-guessing campaign, with more than 151,000 failed sign-ins against 327 accounts.
  • Storm-3358 used ARTEX agents to exploit anonymous file-upload flaws and flaws that caused servers to send requests for the attacker, then installed open-source remote-access tools.
  • Crafty Falcon reportedly entered through internet-facing test and staging systems and used credentials or tokens shared with production, stolen long-lived API tokens, and forged repository commits.
  • Storm-0940 stole credentials for ConnectWise Cloud Backup and used ARTEX to attempt sign-ins to more than 2,250 customer tenants.
  • A modified ARTEX Docker image contained a back door that ran commands submitted through the login form as root.
  • ARTEX consoles were exposed to the internet on port 8787 when deployed with the supplied Docker Compose configuration.
Defensive Notes
  • Scan public and internal repositories, including Git history, for secrets; rotate every exposed secret even if the file or commit was deleted.
  • Replace long-lived cloud keys in CI with short-lived OIDC sign-in tokens where possible, and set expirations on personal access tokens.
  • Check internet-facing systems, especially test and staging environments, for default, shared, or development credentials; do not share credentials between test and production.
  • Review which ARTEX copy and Docker image are installed. Rebuild hosts using tomicdd/artex or congdonglinux/artex and rotate credentials stored on them.
  • Keep ARTEX consoles off the internet, bind port 8787 to localhost, and enforce target scope with network controls.
  • Search web, proxy, and API gateway logs for the listed ARTEX indicators and request patterns; review Entra ID sign-in errors AADSTS7000215 and AADSTS7000222.
  • Investigate servers or CI runners making unexplained AI model API calls or unexpected outbound network connections.

Indicators of compromise

TypeIndicatorContext
IPV4144[.]172[.]64[.]128Storm-0940 infrastructure identified by Microsoft and Wiz.
IPV4144[.]172[.]64[.]60Storm-0940 infrastructure identified by Microsoft and Wiz.
IPV4144[.]172[.]64[.]99Storm-0940 infrastructure identified by Microsoft and Wiz.
IPV4157[.]22[.]126[.]128Microsoft identified this IP in exploit attempts and reconnaissance against Azure AI workloads.
IPV4178[.]16[.]54[.]129ARTEX server infrastructure attributed to Storm-3358 by Microsoft.
IPV4199[.]19[.]77[.]84IP associated by Microsoft with application sign-ins using credentials found on GitHub.
IPV431[.]77[.]229[.]35ARTEX server infrastructure attributed to Storm-3358 by Microsoft.
IPV438[.]244[.]50[.]120ARTEX server associated with the South Korean finance attacks, according to CrowdStrike.
IPV465[.]87[.]223[.]26Microsoft identified this IP in exploit attempts and reconnaissance against Azure AI workloads.
IPV485[.]11[.]168[.]45ARTEX server infrastructure attributed to Storm-3358 by Microsoft.
IPV491[.]92[.]240[.]238ARTEX server infrastructure attributed to Storm-3358 by Microsoft.

MITRE ATT&CK

People

Threat Actors

Vendors

Products

Tools

Countries

Industries

Related Articles