EXPOSURE vLLM
SEVERITY HIGH
PROBE GET :8000/v1/models
Is your vLLM exposed?
Unauthenticated inference endpoint: free compute for whoever finds it first.
Scan a host you own → Fix card Census
Why it matters
Your model-serving API answers anyone: strangers can list your models and run unlimited inference on your GPU — you pay the compute bill, and they can read anything the model was prompted with through your app.
How common is it?
311 vLLM instances are reachable in the Shodan index. Fingerprint match, not a verified zero-auth count · src: Shodan count · `http.html:"vLLM"` · 2026-08
Counts are population-level, from third-party indexes and my own honeypots. I never scan the internet and this page makes no claim about any specific host.
Fix it
The full walkthrough lives on the fix card, kept current in one place: