Suspicious AI Agent Compromise via Web Worker Spawning Shell or Network Tool

PremiumReviewedSigma · Medium · v1
Product
linux
Category
process_creation
Author
HuntRule
Published
2026-09-19
Updated
2026-09-19

ATT&CK techniques

Initial Access → Execution
  1. Recon

  2. Resource Dev

  3. Persistence

  4. Priv Esc

  5. Defense Evasion

  6. Cred Access

  7. Discovery

  8. Lateral Movement

  9. Collection

  10. C2

  11. Exfiltration

  12. Impact

What it detects

This rule detects an application worker such as gunicorn, uvicorn, celery or a python or node runtime spawning curl, wget or a shell which mirrors the AI agent abuse seen in the Hugging Face breach. Attackers who reach an exposed AI service pivot to shell and network utilities to pull tooling and exfiltrate data. Web worker parents launching these processes warrant investigation for server-side compromise.

Related detections9 linkedT1059.004 — drag to rearrange
Possible TP-Link Command Injection via WlanNetworkRpm Endpoint (via webserver)
Malicious Command Injection Probe via node Spawning Shell Arithmetic Echo (via process_creation)
Malicious Reverse Shell via bash dev tcp Redirection (via process_creation)
Suspicious GeoServer Java Process Spawning Download Utility
Possible Shellshock Bash Function Definition Injection via Process Creation
Malicious React Server Component RCE via Node Web Process Spawning Shell (via process_creation)
Malicious Reverse Shell Spawned from Oracle EBS Java Process
Suspicious NKAbuse Implant Staging under StoreService Config Directory (via file_event)
Malicious n8n Expression Sandbox Escape child_process Payload
Suspicious AI Agent Compromise via Web Worker Spawning Shell or Network Tool
Pivot detection · T1059.004 · 9 related

Detection content is published as a reviewed draft. Tune thresholds and exclusions against your own telemetry before enabling this rule for alerting.