ARTEX AI Pentest Tool Found on Servers Attacking Japanese Companies

Summary

Attackers are using ARTEX, an open-source agentic AI penetration testing tool developed in China, to automate attacks against organizations in Japan and South Korea. The framework leverages Large Language Models to perform reconnaissance, probing, and exploitation, helping attackers uncover “shadow vulnerabilities” and evade traditional WAF defenses. The observed campaigns appear highly automated and focused on specific industries.

Investigation

Investigations by GMO Cybersecurity, CrowdStrike, and JPCERT/CC uncovered AI agent configurations and session histories on attacker-controlled infrastructure. Forensic analysis revealed the use of ARTEX, Claude Code, and multiple LLM API connections, including DeepSeek, Zhipu AI, and xAI. Chinese-language prompts and session logs further confirmed that generative AI was actively used to support the intrusion process.

Mitigation

Organizations should implement a Software Bill of Materials (SBOM) to maintain accurate visibility into software versions and deployed components. Security teams should move beyond version-based patching and adopt policies that prioritize immediate upgrades to fixed releases after vulnerabilities are disclosed. WAF protections should also be reinforced with rate limiting tied to accounts, sessions, and API keys rather than IP addresses alone.

Response

When unusually high volumes of automated requests or distributed probing activity are detected, incident responders should investigate for AI-driven reconnaissance patterns. Organizations should review API authorization models to ensure enforcement occurs on the server side and audit exposed administrative functions. If a peer organization in the same industry reports a compromise, proactively scan exposed APIs and third-party platforms for similar weaknesses.

Simulation Execution

• Attack Narrative & Commands: An adversary has gained initial access to a Japanese corporate workstation. To accelerate their reconnaissance and exploitation phase, they deploy the ARTEX AI-driven penetration tool. The attacker executes the tool via the command line to initiate an automated scanning phase against the internal network. The command ARTEX.exe --scan --target 192.168.1.0/24 is used, which is expected to trigger the detection rule due to the presence of the ‘ARTEX’ string in the process creation telemetry.

• Regression Test Script:

  # Simulation Script: Triggering the ARTEX detection rule
  # This script simulates the execution of the ARTEX tool by creating a dummy executable
  # and running it with the targeted command-line argument.
  
  $SimulatedToolPath = "$env:TEMPARTEX.exe"
  
  # Create a dummy executable (using a simple PowerShell wrapper for simulation purposes)
  # In a real scenario, this would be the actual ARTEX binary.
  New-Item -Path $SimulatedToolPath -ItemType File -Force | Out-Null
  
  Write-Host "[+] Simulating ARTEX tool execution..." -ForegroundColor Cyan
  
  # Execute the 'tool' with the specific string required by the detection logic
  # We use cmd /c to ensure the command line string is captured clearly in the logs
  Start-Process cmd.exe -ArgumentList "/c $SimulatedToolPath --scan --target 10.0.0.1" -Wait
  
  Write-Host "[+] Simulation command sent. Check SIEM for EventID 4688 containing 'ARTEX'." -ForegroundColor Green

• Cleanup Commands:

  # Cleanup Script: Remove the simulated artifact
  $SimulatedToolPath = "$env:TEMPARTEX.exe"
  if (Test-Path $SimulatedToolPath) {
      Remove-Item -Path $SimulatedToolPath -Force
      Write-Host "[+] Cleanup complete: $SimulatedToolPath removed." -ForegroundColor Yellow
  } else {
      Write-Host "[!] Cleanup skipped: Artifact not found." -ForegroundColor Red
  }