← Voltar ao feed
medium

ARTEX-derived AI agent ScopeWeaver automates enterprise intrusion attempts with Nmap, PsExec and WinRM in ExtraHop tests

1 fonte(s) · 1 atualização(ões) · publicado 16/11/2026 02:57 · coletado há 1 hora

Reveal Labs (ExtraHop) tested ScopeWeaver, a localized fork of the ARTEX AI offensive framework, in controlled simulations. The agent chained Nmap, PsExec, Evil-WinRM and WinRM attempts against Windows hosts and switched techniques after failures. It generated detectable network activity even when it failed. The article says ARTEX was released for authorized penetration testing in July 2026 and remains available through forks, although its developer removed the original repository on October 8. It also cites a November 13 FBI FLASH warning about machine-speed intrusions. No independent research was performed, so this summary rests on the article alone. The tests used supplied credentials, so they do not show an unaided compromise.

Detalhes técnicos

  • Architecture: a planner turns operator objectives into intents. Concurrent workers run tools and store findings in PostgreSQL, which feeds later planning.
  • Test 1: SSH and RDP failed. Nmap found WinRM ports 5985 and 5986. PsExec, RDP clients, Evil-WinRM and Test-WSMan followed. No shell was obtained, yet the agent declared success, mistaking service reachability for exploitation.
  • Test 2: a shell was established through Impacket PsExec, and the NoLockScreen registry value was set to 1. No persistence was set.
  • Detections: TCP SYN scan, scanner user agents, dual-use software, remote service launch, PowerShell remoting, plaintext HTTP authentication, Impacket PsExec and NetExec activity.
  • Artifacts: user agents "scopeweaver", "artex-enrich/1.0" and "Mozilla/5.0 (spa-api-recon)". LLM API function names list_findings, record_fact and shell_send. None of these alone proves ARTEX activity, and visibility of them needs TLS decryption.
  • Context: Censys reported 1,465 exposed consoles, not necessarily malicious. Microsoft documented ARTEX-assisted authentication attempts against more than 2,250 Microsoft 365 tenants using compromised ConnectWise application credentials. Google observed execution bursts of about 2 to 25 seconds.

Remediação

Correlate attack sequences (scanning followed by exploitation, technique changes, first-time outbound connections) rather than trying to spot an AI operator. Test user-agent or JA4 signatures against historical traffic for false positives. Keep human validation of agent-reported success.

Linha do tempo

  1. gbhackers.com ↗ · pub. 16/11/2026 02:57 · coletado 16/11 04:00 primeiro registro

    ARTEX-derived AI agent ScopeWeaver automates enterprise intrusion attempts with Nmap, PsExec and WinRM in ExtraHop tests

    ler a análise desta fonte

    Reveal Labs (ExtraHop) tested ScopeWeaver, a localized fork of the ARTEX AI offensive framework, in controlled simulations. The agent chained Nmap, PsExec, Evil-WinRM and WinRM attempts against Windows hosts and switched techniques after failures. It generated detectable network activity even when it failed. The article says ARTEX was released for authorized penetration testing in July 2026 and remains available through forks, although its developer removed the original repository on October 8. It also cites a November 13 FBI FLASH warning about machine-speed intrusions. No independent research was performed, so this summary rests on the article alone. The tests used supplied credentials, so they do not show an unaided compromise.

    Technical Details

    • Architecture: a planner turns operator objectives into intents. Concurrent workers run tools and store findings in PostgreSQL, which feeds later planning.
    • Test 1: SSH and RDP failed. Nmap found WinRM ports 5985 and 5986. PsExec, RDP clients, Evil-WinRM and Test-WSMan followed. No shell was obtained, yet the agent declared success, mistaking service reachability for exploitation.
    • Test 2: a shell was established through Impacket PsExec, and the NoLockScreen registry value was set to 1. No persistence was set.
    • Detections: TCP SYN scan, scanner user agents, dual-use software, remote service launch, PowerShell remoting, plaintext HTTP authentication, Impacket PsExec and NetExec activity.
    • Artifacts: user agents "scopeweaver", "artex-enrich/1.0" and "Mozilla/5.0 (spa-api-recon)". LLM API function names list_findings, record_fact and shell_send. None of these alone proves ARTEX activity, and visibility of them needs TLS decryption.
    • Context: Censys reported 1,465 exposed consoles, not necessarily malicious. Microsoft documented ARTEX-assisted authentication attempts against more than 2,250 Microsoft 365 tenants using compromised ConnectWise application credentials. Google observed execution bursts of about 2 to 25 seconds.

    Remediation

    Correlate attack sequences (scanning followed by exploitation, technique changes, first-time outbound connections) rather than trying to spot an AI operator. Test user-agent or JA4 signatures against historical traffic for false positives. Keep human validation of agent-reported success.