# ARTEX open-source agentic AI pentest framework abused by threat actors (Storm-3358, Storm-0940, others) for automated recon and exploitation

> ARTEX, an open-source Go-based agentic AI penetration-testing framework published on GitHub on July 26, 2026, has been weaponized by multiple clusters, including Microsoft-tracked Storm-3358 (financial organizations) and Storm-0940 (ConnectWise Cloud Backup supply chain, IT/software vendors), a Chinese-speaking actor that breached South Korean banks, and operators attacking Japanese companies. The FBI issued a FLASH advisory in November 2026.

- **Published:** 2026-11-17T14:35:37Z
- **Last reviewed:** 2026-11-17T14:35:37Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6252
- **Card image:** https://intel.threadlinqs.com/og/threat/TL-2026-6252.png
- **ID:** TL-2026-6252
- **Severity:** HIGH
- **Category:** THREAT_INTEL
- **Status:** ACTIVE
- **Actor:** Storm-3358 (China)
- **Detections:** 9 (1 shown free below) · **IOCs:** 26 (3 shown free below)
- **CVEs:** CVE-2026-109618, CVE-2026-109619

## Description

ARTEX (GitHub: Autumn-27/ARTEX, AGPL-3.0, released 2026-07-26) is an LLM-driven autonomous pentesting framework written in Go with a Next.js console on PostgreSQL, bundling nmap and mitmproxy and syncing assets from the same author's ScopeSentry platform. A Planner agent converts an operator goal (e.g. 'gain admin panel access') into attack intents; Worker agents claim intents, run tools such as nmap, and record results in an asset graph and an exploration graph until the goal is met. It has no model of its own and drives Anthropic, OpenAI or any OpenAI-compatible endpoint; observed abuse used DeepSeek, GLM, Grok, Kimi, Zhipu AI and xAI models, frequently through self-hosted LLM proxies/resellers (CLIProxyAPI, New API) and the reseller domain xcai.pro. Mandiant/GTIG recorded 27 events in 23 seconds and automated error correction used to land webshells. The default console listens on tcp/8787.

CrowdStrike (2026-10-07) reported a likely Chinese-speaking, financially motivated, unattributed actor running ARTEX from 38.244.50.120 (open directory on port 18899; DeepSeek primary model plus Claude Code sessions backed by GLM and Grok) against South Korean finance; a broker-facing loan service and an employee mobile system were breached and data exfiltrated (FBI/press cite Shinhan, KB Kookmin and Hana, ~67,000 people). The author disowned malicious use and removed the repository on 2026-10-08; forks and rebrands (e.g. ScopeWeaver) and a commercial ARTEX PRO persist. Microsoft attributed an ARTEX-driven ConnectWise Cloud Backup supply-chain compromise to China-state-aligned Storm-0940 (sign-in attempts against 2,250+ Microsoft 365 tenants peaking at 2,900/min; privileged cloud access in under three hours in some cases), followed by Wiz-reported Storm-0940 activity against IT/software vendors with Okta, Entra ID and AWS Cognito authentication probing, account-recovery abuse and IP/user-agent rotation. Microsoft (2026-11-06) described Storm-3358, a financially motivated Cambodia-based cluster (per secondary coverage), using ARTEX with conventional tools to find SSRF and anonymous file-upload flaws in internet-facing systems of financial organizations (mainly Southeast Asia), then deploying WISP, vshell and AdaptixC2 RATs and abusing cloud identities. Google/GTIG tracks a related campaign as UNC7228 (overlap with Storm-0940 per secondary reporting); CrowdStrike separately ties VAPOR PANDA to ARTEX infrastructure (secondary reporting, unverified). ARTEX was also seen on servers attacking Japanese companies (API probing and WAF evasion in Japan and South Korea).

Censys counted 1,465 exposed ARTEX consoles on Nov 3-4, 2026 (US 726, Hong Kong 387, China 177, Japan 51), concentrated in PEG Tech AS54600 (493 hosts) and Yancy Limited AS138415 (157), with 32 hosts exposing open directories containing post-exploitation kits (Cobalt Strike, AdaptixC2, Vshell, FRP, NPS, GodPotato, fscan, stagers). Related agentic frameworks (CyberStrikeAI, PentAGI, HexStrike AI, CAIRN) appear on the same hosts. The related PHP2Shell flaws (CVE-2026-109618, CVE-2026-109619; CVSS v4.0 9.2; unauthenticated RCE in PHP core; fixed in PHP 8.5.12/8.4.27/8.3.36/8.2.35 on 2026-10-21) have public PoCs and the SOC Prime source describes them as the bug class agentic frameworks favor; however, sources disagree on mechanics (SOC Prime: session.upload_progress chain affecting PHP 5.4.0-8.5.11; Arctic Wolf: out-of-memory memory corruption and stale request-state) and Arctic Wolf states in-the-wild exploitation success is not firmly established and no direct ARTEX-to-PHP2Shell link is proven. No CVSS score applies to ARTEX itself. Behavior-based detection is recommended over name-based, since IOCs are perishable.

## MITRE ATT&CK

- T1595.002 Vulnerability Scanning
- T1588.002 Tool
- T1588.007 Artificial Intelligence
- T1190 Exploit Public-Facing Application
- T1078.004 Cloud Accounts
- T1110.003 Password Spraying
- T1059.004 Unix Shell
- T1505.003 Web Shell
- T1098 Account Manipulation
- T1046 Network Service Discovery
- T1018 Remote System Discovery
- T1580 Cloud Infrastructure Discovery
- T1071.001 Web Protocols
- T1571 Non-Standard Port
- T1219 Remote Access Tools
- T1090 Proxy

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

**ARTEX Post-Exploitation: Web Server Process Spawning Shell or Recon Tools (Linux)** (`TL-2026-6252-SPL-002`)

Detects Linux web-server and PHP runtime processes (php-fpm, php-cgi, apache2, httpd, nginx, lsphp, FrankenPHP) spawning shells, nmap, curl, wget or interpreters, the behaviour seen after ARTEX agents land webshells via SSRF/file-upload or PHP2Shell flaws. Behaviour over tool names, per the research recommendation (service accounts spawning shells, nmap/curl/wget from unexpected parents).

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1059.004, T1505.003, T1046, T1018
- **Data sources:** splunk:index=linux:sourcetype=sysmon:linux, microsoft_defender:DeviceProcessEvents, sysmon_for_linux
- **Severity:** high
- **Confidence:** high

### Splunk SPL

````spl
index=linux sourcetype="sysmon:linux" (EventCode=1 OR EventID=1) earliest=-24h
```TL-2026-6252 ARTEX webshell post-exploitation T1059.004 T1505.003 T1046 T1018```
| eval parent_base=lower(mvindex(split(ParentImage,"/"),-1)), img_base=lower(mvindex(split(Image,"/"),-1))
| where match(parent_base,"^(php-fpm[0-9.]*|php-cgi|lsphp[0-9]*|php[0-9.]*|apache2|httpd|nginx|frankenphp|lshttpd)$")
| where match(img_base,"^(sh|bash|dash|zsh|nc|ncat|socat|nmap|masscan|curl|wget|python[0-9.]*|perl|ruby)$")
| table _time host User ParentImage ParentCommandLine Image CommandLine
| sort 0 - _time
````

### Microsoft KQL

```kql
// TL-2026-6252 ARTEX web server spawning shell or recon tooling
let WebParents = dynamic(["php-fpm","php-cgi","php","lsphp","apache2","httpd","nginx","frankenphp","lshttpd","w3wp.exe","php-cgi.exe","httpd.exe"]);
let SpawnedTools = dynamic(["sh","bash","dash","zsh","nc","ncat","socat","nmap","masscan","curl","wget","python3","perl","cmd.exe","powershell.exe","whoami.exe"]);
DeviceProcessEvents
| where Timestamp > ago(24h)
| where tolower(InitiatingProcessFileName) in (WebParents)
| where tolower(FileName) in (SpawnedTools)
| project Timestamp, DeviceName, AccountName, InitiatingProcessFileName, InitiatingProcessCommandLine, FileName, ProcessCommandLine
| order by Timestamp desc
```

### Sigma

```yaml
title: Web Server or PHP Runtime Spawning Shell or Recon Tool (ARTEX Webshell Activity)
id: 2b9e4a7c-51d3-4f68-a3e0-7c18d94f2b52
status: experimental
description: Detects Linux web server and PHP processes spawning shells, nmap, curl, wget or interpreters, consistent with ARTEX agent post-exploitation after webshell deployment.
references:
  - https://www.crowdstrike.com/en-us/blog/unknown-threat-actor-uses-artex-to-target-south-korean-finance/
  - https://socprime.com/blog/artex-detection-hunting-the-ai-pentest-framework-turned-attack-platform/
author: Threadlinqs Team
date: 2026-11-17
tags:
  - attack.execution
  - attack.t1059.004
  - attack.persistence
  - attack.t1505.003
  - attack.discovery
  - attack.t1046
  - attack.t1018
logsource:
  category: process_creation
  product: linux
detection:
  selection_parent_contains:
    ParentImage|contains:
      - 'php-fpm'
      - 'php-cgi'
      - 'lsphp'
      - 'frankenphp'
  selection_parent_endswith:
    ParentImage|endswith:
      - '/apache2'
      - '/httpd'
      - '/nginx'
      - '/php'
  selection_child:
    Image|endswith:
      - '/sh'
      - '/bash'
      - '/dash'
      - '/nc'
      - '/ncat'
      - '/socat'
      - '/nmap'
      - '/curl'
      - '/wget'
      - '/python3'
      - '/perl'
  condition: 1 of selection_parent_* and selection_child
falsepositives:
  - Web applications that legitimately shell out (backup, image processing, mail)
  - Administrators running commands under a web service context
level: high
```

**Known false positives:**

- CMS or application plugins that legitimately invoke curl, wget or shell helpers from php-fpm
- Administrators troubleshooting under a web service account

**Tuning notes:** Baseline legitimate web-to-shell parent/child pairs per host and exclude by CommandLine; alert higher when child is nmap, nc or socat or when the command line contains an external IP; extend parent list for Tomcat or Node if used.

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

- Storm-0940 ARTEX-Driven Cloud Identity Password Spray with Successful Sign-In (locked: unlock with TL Creds)
- ARTEX Agent Cloud Permission Testing via AWS EC2 Dry-Run Burst (locked: unlock with TL Creds)
- ARTEX Operator LLM Reseller and Model-API DNS Lookups from Hosts (locked: unlock with TL Creds)
- Web Server Process Writing Script File to Web Root (ARTEX Webshell Upload) (locked: unlock with TL Creds)
- ARTEX Post-Exploitation Toolkit Execution (GodPotato, fscan, FRP, vshell, AdaptixC2, Invoke-TheHash) (locked: unlock with TL Creds)
- Payload Download from ARTEX / PHP2Shell Staging Infrastructure via curl, wget or LOLBin (locked: unlock with TL Creds)
- ARTEX API Recon and PHP2Shell Session-Upload Exploitation in Web Logs (locked: unlock with TL Creds)
- ARTEX Ecosystem C2 and Staging Egress to Known IOCs and Operator Ports (locked: unlock with TL Creds)

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

## Sample indicators (3 of 26)

Defanged, TLP:CLEAR.

- IP address: `38.244.50[.]120`
- Domain: `xcai[.]pro`
- URL: `hxxp://45[.]115[.]26[.]26:8000/cCmono`

Full set of 26 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [ARTEX Detection: Hunting the AI Pentest Framework Turned Attack Platform (SOC Prime)](https://socprime.com/blog/artex-detection-hunting-the-ai-pentest-framework-turned-attack-platform/)
- [Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance (CrowdStrike)](https://www.crowdstrike.com/en-us/blog/unknown-threat-actor-uses-artex-to-target-south-korean-finance/)
- [Censys: ARTEX agentic asbestos](https://censys.com/blog/artex-agentic-asbestos/)
- [FBI warns ARTEX AI hacking tool is powering global cyberattacks (BleepingComputer)](https://www.bleepingcomputer.com/news/security/fbi-warns-artex-ai-hacking-tool-is-powering-global-cyberattacks/)
- [Storm-3358 / ARTEX abuse expands to financial organizations (SOC Prime)](https://socprime.com/active-threats/artex-abuse-expands-as-cybercriminals-target-financial-organizations/)
- [Storm-0940 broadens ARTEX attacks to IT and software vendors (SOC Prime)](https://socprime.com/active-threats/storm-0940-broadens-artex-attacks-to-it-and-software-vendors/)
- [ARTEX leads a new wave of agentic offensive frameworks (SOC Prime)](https://socprime.com/active-threats/artex-leads-a-new-wave-of-agentic-offensive-frameworks/)
- [ARTEX AI pentest tool found on servers attacking Japanese companies (SOC Prime)](https://socprime.com/active-threats/artex-ai-pentest-tool-found-on-servers-attacking-japanese-companies/)
- [PHP2Shell: critical PHP vulnerabilities enable unauthenticated RCE (Arctic Wolf)](https://arcticwolf.com/resources/blog/php2shell-critical-php-vulnerabilities-enable-unauthenticated-remote-code-execution/)
- [PHP announces critical security release for two vulnerabilities (Qualys)](https://blog.qualys.com/vulnerabilities-threat-research/2026/10/20/php-announces-critical-security-release-for-two-vulnerabilities)
- [China-linked hackers use ARTEX AI (The Hacker News)](https://thehackernews.com/2026/10/china-linked-hackers-use-artex-ai.html)
- [Multi-Agent Offensive Frameworks in the Wild: CyberStrikeAI and ARTEX (Cloud Security Alliance)](https://labs.cloudsecurityalliance.org/research/csa-research-note-cyberstrikeai-artex-multi-agent-offense-20/)
- [PHP2Shell bugs signal red alert for security industry (Dark Reading)](https://www.darkreading.com/application-security/php2shell-bugs-red-alert-security-industry)
- [Microsoft Reports Storm-3358 Using ARTEX Against Financial Sector (secondary coverage of Microsoft Security Blog, 2026-11-06)](https://letsdatascience.com/news/microsoft-reports-storm-3358-using-artex-against-financial-s-549e54ae)

## 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): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6252
