# ARTEX AI-Powered Offensive Framework Abused by Criminal and State-Linked Actors to Automate Attacks Without Zero-Days (Storm-3358, UNC7228/Storm-0940, Trojanized Docker Images)

> ARTEX, an open-source agentic LLM pentesting/offensive-C2 framework, is being abused by financially motivated and suspected PRC-nexus actors (Microsoft Storm-3358, Storm-0940 / GTIG UNC7228, CrowdStrike Vapor Panda) to automate attacks on ordinary weaknesses (anonymous file upload, SSRF, SQL injection, weak credentials, over-permissive cloud identities) and reach privileged access in under three hours. Trojanized third-party Docker images (tomicdd/artex, congdonglinux/artex) additionally expose operators' consoles to unauthenticated root command execution.

- **Published:** 2026-11-20T11:49:24Z
- **Last reviewed:** 2026-11-20T11:49:24Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6572
- **Card image:** https://intel.threadlinqs.com/og/threat/TL-2026-6572.png
- **ID:** TL-2026-6572
- **Severity:** HIGH
- **Category:** THREAT_INTEL
- **Status:** ACTIVE
- **Actor:** Storm-3358
- **Detections:** 9 (1 shown free below) · **IOCs:** 28 (3 shown free below)

## Description

ARTEX is an autonomous multi-agent LLM pentesting and offensive command-and-control platform written in Go with an embedded Next.js web console (GTIG, 2026-10-30). The model proposes an action, the framework executes it with common tooling (subfinder, httpx, Nmap, nuclei, ffuf, curl/wget, Python HTTP clients, Playwright/headless Chromium), and the output feeds the next decision. The framework holds model API keys, target data and captured sessions. The original developer's repository was taken down in October 2026, but forks (ScopeWeaver, a commercial 'ARTEX PRO', and many copycat GitHub/Docker Hub namespaces) continue to circulate. More than 1,600 ARTEX consoles were internet-exposed by early November 2026 (BadByte).

Abuse is multi-actor. Microsoft (2026-11-06) tracks Storm-3358, a financially motivated, Cambodia-based cluster targeting financial organizations since October 2026, mainly in Southeast Asia, which pairs ARTEX with conventional scanners to exploit anonymous file upload and SSRF flaws and installs the open-source RATs WISP, vshell and AdaptixC2 (downloaded with netcat/curl) for persistence and lateral movement. Google Threat Intelligence Group / Mandiant (2026-10-30) document ARTEX in intrusion investigations: rapid AWS permission probing, MySQL general-log poisoning to write JSP web shells, Kudu/SCM command execution, Kerberoasting and DCSync, enumeration of over 110,000 AD objects, command output posted to rotating webhook.site URLs, and tight execution bursts (e.g., 27 events in 23 seconds). GTIG tracks the Wiz-reported campaign as UNC7228, overlapping Microsoft's PRC-nexus Storm-0940 (also reported targeting M365 tenants via compromised ConnectWise application credentials). CrowdStrike reporting cited by BadByte covers 'Vapor Panda' and ARTEX use against South Korean banks; CERT-In tracks an AI-assisted actor 'Crafty Falcon' against Indian BFSI. An FBI FLASH (reported 2026-11-13 by Cyber Security News; primary PDF not independently verified) says abuse has occurred since at least August 2026 and that some intrusions reached privileged access in under three hours. Shared ARTEX use does not by itself prove these actors are connected.

Supply-chain dimension: BadByte (2026-11-04/05) found the third-party Docker images tomicdd/artex (account created 2026-10-08; ~9,131 pulls) and congdonglinux/artex (~45 pulls) contain a trojanized /app/artex binary. A new function server.verifyCaptcha passes a secret marker in a 'captcha' field of POST /api/auth/login to sh -c as root, before credential checks, returning command output in the response. The implant does not beacon; it waits for requests. 62 backdoored consoles were observed Nov 2-4 (26 US, 17 Hong Kong, 7 China). Exposed secrets include LLM API keys, JWT signing key, MCP tokens and webhook/SMTP secrets. Docker removed both images on 2026-11-05.

A benign-but-instructive incident: WorkOS disclosed that a researcher running ARTEX benchmark evaluations had an agent reach an internal sales/support tool (public API, predictable development credential, no rate limiting, telemetry gaps) between Oct 25 and Oct 30, 2026; WorkOS attributes it to non-malicious activity. WorkOS published user-agent and JA4 indicators.

Defensive takeaways: ARTEX does not need zero-days. Defenders should harden exposed apps (file upload, SSRF), enforce least-privilege cloud identities, remove weak/predictable credentials, add rate limiting, monitor for high-volume varied requests, egress to webhook/OAST services and curl-driven downloads, and use decoys/tarpits to waste agent effort. Sparse primary IOCs mean behavioral detection is essential.

## MITRE ATT&CK

- T1588.007 Obtain Capabilities: Artificial Intelligence
- T1595.002 Active Scanning: Vulnerability Scanning
- T1046 Network Service Discovery
- T1190 Exploit Public-Facing Application
- T1078 Valid Accounts
- T1195.002 Supply Chain Compromise: Compromise Software Supply Chain
- T1059.004 Command and Scripting Interpreter: Unix Shell
- T1505.003 Server Software Component: Web Shell
- T1136 Create Account
- T1098 Account Manipulation
- T1110.003 Brute Force: Password Spraying
- T1558.003 Steal or Forge Kerberos Tickets: Kerberoasting
- T1003.006 OS Credential Dumping: DCSync
- T1087.002 Account Discovery: Domain Account
- T1580 Cloud Infrastructure Discovery
- T1071.001 Application Layer Protocol: Web Protocols
- T1567 Exfiltration Over Web Service

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**ARTEX Trojanized Console Root Backdoor — Shell Spawned by /app/artex (verifyCaptcha)** (`TL-2026-6572-SPL-001`)

Detects the backdoor in trojanized ARTEX Docker images tomicdd/artex and congdonglinux/artex (SHA-256 d5070f52262effe216b840711343bff8f2787f39208d887d8dfd9c673f8d1573): the /app/artex Go binary passes a secret marker in the captcha field of POST /api/auth/login to sh -c as root before credential checks. The implant does not beacon, so the only host signal is a shell spawned by the ARTEX process. Exposed consoles default to port 8787.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1195.002, T1059.004, T1190
- **Data sources:** sysmon_for_linux, linux_auditd, microsoft_defender_for_endpoint_linux
- **Severity:** critical
- **Confidence:** high

### Splunk SPL

````spl
```comment: TL-2026-6572 ARTEX trojanized console backdoor - tomicdd/artex and congdonglinux/artex run sh -c as root from /app/artex```
index=linux sourcetype="sysmon:linux" EventID=1 earliest=-24h
(ParentImage="/app/artex" OR ParentImage="*/artex")
(Image="/bin/sh" OR Image="/usr/bin/sh" OR Image="/bin/dash" OR Image="/bin/bash" OR Image="/usr/bin/bash")
| where match(CommandLine,"(^|\s)-c(\s|$)")
| stats count min(_time) as first_seen max(_time) as last_seen values(CommandLine) as commands values(User) as users by host ParentImage Image
| convert ctime(first_seen) ctime(last_seen)
````

### Microsoft KQL

```kql
let ArtexSha = "d5070f52262effe216b840711343bff8f2787f39208d887d8dfd9c673f8d1573";
union
(DeviceProcessEvents
 | where Timestamp > ago(24h)
 | where InitiatingProcessFolderPath =~ "/app/artex" or InitiatingProcessFileName =~ "artex"
 | where FileName in~ ("sh", "bash", "dash") and ProcessCommandLine contains " -c"
 | extend Signal = "shell_spawned_by_artex"),
(DeviceFileEvents
 | where Timestamp > ago(24h)
 | where SHA256 =~ ArtexSha
 | extend Signal = "trojanized_artex_binary_hash")
| project Timestamp, DeviceName, AccountName, Signal, FileName, ProcessCommandLine, InitiatingProcessFolderPath, InitiatingProcessCommandLine, SHA256
| order by Timestamp desc
```

### Sigma

```yaml
title: ARTEX Trojanized Console Spawns Root Shell (verifyCaptcha Backdoor)
id: 7c1f3a52-4d8e-4b6a-9a31-2f5e8c0d1a01
status: experimental
description: Detects a shell launched with -c by the /app/artex binary from the trojanized tomicdd/artex and congdonglinux/artex Docker images (POST /api/auth/login captcha field backdoor).
references:
    - https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/
    - https://cloud.google.com/blog/topics/threat-intelligence/artex-and-friends-enter-agentic-offensive-frameworks
    - https://badbyte.io/artex-backdoor-docker-image/
author: Threadlinqs Team
date: 2026/11/20
tags:
    - attack.initial-access
    - attack.t1195.002
    - attack.execution
    - attack.t1059.004
    - attack.t1190
logsource:
    category: process_creation
    product: linux
detection:
    selection_parent:
        ParentImage|endswith: '/artex'
    selection_shell:
        Image|endswith:
            - '/sh'
            - '/bash'
            - '/dash'
        CommandLine|contains: ' -c'
    condition: selection_parent and selection_shell
falsepositives:
    - Legitimate ARTEX operators whose own build shells out for tool execution (verify the image digest against reviewed source)
level: critical

```

**Known false positives:**

- Legitimate, reviewed ARTEX builds that shell out via sh -c to run pentest tooling (nmap, nuclei, curl)
- Authorized red-team hosts running official ARTEX builds

**Tuning notes:** Allow-list hosts known to run the official ARTEX build by verified image digest. Keep the alert on any internet-exposed ARTEX console (port 8787). Pair with HTTP body inspection for POST /api/auth/login containing a captcha field where a WAF or Stream is available. On a hit, rotate LLM API keys, jwt.key, MCP tokens and webhook/SMTP secrets.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- ARTEX Agent Web Reconnaissance and Exploitation — User-Agent, Request Burst, Upload and SSRF Probes (locked: unlock with TL Creds)
- ARTEX Active Directory Compromise — Kerberoasting, DCSync and Bulk LDAP Enumeration (locked: unlock with TL Creds)
- ARTEX Command Output Exfiltration to webhook.site and LLM API Reseller (locked: unlock with TL Creds)
- Trojanized ARTEX Docker Image Pull or Run ([indicator], [indicator]) (locked: unlock with TL Creds)
- MySQL General-Log Poisoning — JSP Web Shell Written by Database Process (locked: unlock with TL Creds)
- Open-Source RAT Download via curl, wget or netcat (WISP, vshell, AdaptixC2) (locked: unlock with TL Creds)
- AWS Rapid Permission Probing Burst — Agentic Discovery with AccessDenied Spike (locked: unlock with TL Creds)
- Storm-3358 ARTEX / WISP / AdaptixC2 Infrastructure Connections (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (3 of 28)

Defanged, TLP:CLEAR.

- IP address: `199.19.95[.]66`
- Domain: `xcai[.]pro`
- File hash: `d5070f52262effe216b840711343bff8f2787f39208d887d8dfd9c673f8d1573`

Full set of 28 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [Hackers Use AI-Powered ARTEX Framework to Automate Cyberattacks Without Relying on Zero-Day Exploits](https://cybersecuritynews.com/hackers-use-ai-powered-artex-framework/)
- [FBI Reveals How Hackers Use ARTEX AI Agents to Gain Privileged Access in Under Three Hours](https://cybersecuritynews.com/fbi-reveals-hackers-artex-ai-usage/)
- [ARTEX abuse expands as cybercriminals target financial sector (Microsoft Threat Intelligence)](https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/)
- [ARTEX and Friends: Enter Agentic Offensive Frameworks (GTIG / Mandiant)](https://cloud.google.com/blog/topics/threat-intelligence/artex-and-friends-enter-agentic-offensive-frameworks)
- [ARTEX Backdoor: Trojanized Docker Image Handing Out Root Shells (BadByte)](https://badbyte.io/artex-backdoor-docker-image/)
- [CERT-In Flags AI-Driven Attacks as Backdoored ARTEX Exposes 62 Consoles to Root Access (The Hacker News)](https://thehackernews.com/2026/11/backdoored-docker-image-for-artex-ai.html)
- [Update for October 30th, 2026 Data Security Incident (WorkOS)](https://workos.com/security/bulletins/update-for-october-30th-2026-data-security-incident)
- [Accuracy Through Volume of Fire (TrustedSec)](https://trustedsec.com/blog/accuracy-through-volume-of-fire)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple and Gold in full, 2 calls a day on a free account): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6572
