HighEcosystem changeOngoing

Artex AI pentest tool abused by attackers, FBI flash alert issued

FBI warns the open-source AI tool Artex is used by criminals and nation-states; scan for leaked secrets and rotate them now.

Published
Updated
Exploitation
Yes, the FBI (as reported by The Record) and StepSecurity, citing vendor research, describe use against many organizations

What happened

Artex (also written ARTEX) is an open-source tool in which AI agents run a penetration test on their own. The operator gives it a target and a goal, and the agents plan the attack, run scanning and exploit tools and record findings, according to StepSecurity.

On Friday, November 13, 2026, the FBI released a flash alert about the tool, according to The Record. The Record quotes the FBI as saying Artex can “automate multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration”, and that some intrusions progressed “from initial targeting to privileged access in under three hours”. We did not read the FBI alert itself; the points below come from The Record’s account of it.

Reported by The Record:

  • The FBI says the framework keeps spreading through community forks, localized variations, archived copies and a commercial version, even though the original GitHub repository was removed.
  • The FBI cites Censys research showing over 1,600 publicly exposed Artex consoles. It says the real number is likely higher once private networks are counted.
  • Artex can use any AI model. The FBI says Chinese models dominate in attacker-controlled instances, with DeepSeek Flash family models most common.
  • The FBI lists targeted sectors: financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecommunications and legal.
  • The FBI says Artex is good at finding less-watched systems, such as partner and supplier portals, mobile backends, employee applications, sales-support systems, legacy services and third-party integrations.
  • Microsoft said Chinese state-backed actors used Artex in a supply chain attack on ConnectWise.
  • An Amazon study, as relayed by The Record, found Chinese-nexus groups using Artex for large-scale theft of credentials for AI model access. It also found one campaign targeting cryptocurrency exchanges for KYC verification bypass using real-time deepfake injection.

Reported by StepSecurity, which summarises research from several vendors and CERTs:

  • Leaked or weak credentials are the common thread. It describes an agent that searched GitHub for secrets and, five minutes later, took data from a customer’s Azure Key Vault, citing Microsoft.
  • In the WorkOS case, an agent from a security researcher’s benchmark tests logged in to an internal sales tool with a guessable development credential that had reached production.
  • After the original repository came down, a copy on GitHub pulled a Docker image with a hidden back door. StepSecurity says users downloaded it more than 9,000 times, and that Docker has since removed the backdoored images.
  • The author posted a notice on October 8 saying attackers had abused the tool, and the repository came down the same day.
  • The supplied Docker Compose file for version 0.3.15 opens the console on port 8787 on every network interface.
  • CERT-In advisory CIAD-2026-0058 describes a group called Crafty Falcon entering through internet-facing test and staging systems, then using credentials and tokens shared with production.

Where the outlets differ:

  • Release date: StepSecurity says Artex appeared on GitHub on July 26, 2026. The Record says August.
  • Exposed consoles: StepSecurity cites a Censys count of 1,465 on November 5. The Record cites over 1,600 in the FBI advisory. The counts may be from different dates; this is not confirmed.
  • Developer: The Record names the developer as Chinese cybersecurity engineer Li Puhua. StepSecurity says no source it read establishes more about the developer than the GitHub account Autumn-27. We could not verify the name independently.

Am I affected?

Not confirmed for any specific organization; the sources give no vendor patch or CVE. Artex is a tool attackers run, not a flaw in one product. Exposure depends on what an automated attacker could reach.

You are more likely to be exposed if:

  • Secrets (cloud keys, tokens, passwords) exist in public or internal repositories or git history. StepSecurity says these are the fastest way in.
  • Internet-facing systems, including test, staging and internal tools, accept development or default credentials. StepSecurity and CERT-In both describe this pattern.
  • You run long-lived cloud keys in CI secrets or personal access tokens with no expiry. StepSecurity lists these as items to replace or limit.
  • Anyone in your organization installed Artex or a fork. StepSecurity says one popular copy shipped a back door.

What to do now

  1. Scan public and internal repositories for secrets, including git history, and rotate every secret you find. StepSecurity notes that deleting a file does not make a leaked secret safe.
  2. Replace long-lived cloud keys in CI secrets with short-lived sign-in tokens (OIDC), and set an expiry on every personal access token (StepSecurity).
  3. Check internet-facing systems, including test, staging, partner portals and internal sales or support tools. None should accept a development or default credential. The FBI, via The Record, lists these lesser-watched systems as common targets.
  4. Make sure test and staging credentials and tokens are not shared with production, since StepSecurity describes Crafty Falcon using exactly that route (per the CERT-In advisory).
  5. Find out whether anyone in your organization runs Artex. Identify which copy and which Docker image they installed, because one popular copy shipped a back door (StepSecurity).
  6. If you run Artex yourself, note that the supplied Docker Compose file exposes port 8787 on all network interfaces (StepSecurity). Do not leave the console internet-facing.
  7. Hunt in logs for the user agents artex-enrich/1.0 and norma/0.4, as suggested by StepSecurity. Its post lists more indicators; the excerpt we read did not include them, so they are not confirmed here.
  8. Expect high-volume attempts. The Record reports the tool generates many attempts that trigger defensive sensors, so review alerting and response speed.

Update, 18 November 2026

  • The Hacker News reports that the backdoored Docker image tomicdd/artex adds a hidden verifyCaptcha() function to the login handler that runs a secret-marked CAPTCHA field through sh -c as root before any credential check, giving an unauthenticated root shell. Researcher Badbyte says this on every install. (thehackernews.com)
  • The Hacker News reports 62 ARTEX consoles served the backdoored build between November 2 and 4, 2026 (26 in the U.S., 17 in Hong Kong, seven in China), and that 1,645 hosts served the ARTEX console as of November 4. Docker removed tomicdd/artex and congdonglinux/artex, but 637 GitHub repositories still point to the backdoored image. (thehackernews.com)
  • CrowdStrike judged with moderate confidence that a single Chinese-speaking, financially motivated lone operator used ARTEX in breaches of seven South Korean financial institutions between late September and early October 2026, affecting roughly 66,000 to 67,000 people and about 2,200 companies, according to Cybersecurity News. (cybersecuritynews.com)
  • Censys’s November 5 report found additional threat detections on 155 ARTEX hosts (10.6%) and 32 open directories exposing tools, implants and operator files. Around 650 of the 1,465 hosts shared repeated configurations, leaving roughly 815 distinct-looking deployments. Censys says public exposure does not by itself prove malicious use. (cybersecuritynews.com)
  • Cybersecurity News, citing Censys, says Microsoft attributed ARTEX-assisted activity against ConnectWise customer tenants to Storm-0940. (cybersecuritynews.com)

Sources