# GCP VM Shielded VM Not Enabled

> Shielded VM protections (Secure Boot, vTPM, integrity monitoring) defend against boot-level rootkits and firmware tampering. ZopNight rule RC-1203 flags enriched Compute Engine instances lacking a shielded_vm_enabled=true stamp, which the discoverer writes when at least 1 of the 3 protections is on. Enabling them requires stopping the VM first.

Source: https://zop.dev/integrations/gcp/recommendations/gcp-vm-shielded-vm-not-enabled
Updated: 2026-08-19

---

## What the shielded_vm_enabled stamp actually attests

Precision matters on this one. The GCP discoverer inspects the instance's `shieldedInstanceConfig` and writes `shielded_vm_enabled=true` when **any** of Secure Boot, vTPM, or integrity monitoring is on. That is an OR, not an AND. Rule RC-1203 then flags enriched instances that lack that `true` stamp, anchoring on the always-present `purchase_type` key so a resource the discoverer never enriched fails safe instead of firing. Two consequences follow. First, the rule infers "not shielded" from the absence of a true-only stamp rather than from an explicit false. Second, an instance with only Secure Boot on (vTPM and integrity monitoring both off) counts as compliant and never fires, a documented producer-side looseness that a rule change cannot tighten.

## Why boot-level integrity is worth a finding

Shielded VM is GCP's defence against the class of attack that survives a rebuild: bootkits and kernel-level rootkits that persist below the OS. Secure Boot refuses unsigned boot components; the vTPM gives the instance measured boot and a sealed key store; integrity monitoring compares each boot's measurements against a baseline and raises a Cloud Monitoring signal when they diverge. None of it protects a running workload from an application-level compromise. The value is knowing the platform underneath was not silently altered.

## Inspect a flagged instance

```bash
gcloud compute instances describe my-vm --zone us-central1-a \
  --format='value(shieldedInstanceConfig)'
```

## Turning it on is a stop-start operation

The Shielded VM options can only be changed while the instance is stopped, so remediation costs a restart window. Order matters, too: enable vTPM and integrity monitoring broadly, but treat Secure Boot with care on VMs that load unsigned kernel modules, since third-party drivers and some older images will refuse to boot with it on. The image must also support UEFI; most current Google-provided images do, imported legacy images may not.

## Preventing new unshielded VMs

Per-instance cleanup does not stop tomorrow's terraform apply from creating another finding. The org policy constraint `compute.requireShieldedVm` blocks non-shielded instance creation across a project, folder, or the organization. Enforce it once and this rule's backlog stops growing.