Skip to content
Free · Checklist

GCP Security Checklist: Score Your Google Cloud

This is an interactive Google Cloud security checklist. Mark 30 controls In place, Partial, Missing or N/A and get a weighted score, a score per domain and your top gaps, each with the Google Cloud documentation behind it.

Answers stay in your browser. For the reference design these controls belong to, read secure GCP architecture; to practise, try the GCP Hardening Lab.

No download, no card

A self-assessment from Google Cloud documentation, not an audit or the CIS benchmark. It does not connect to your Google Cloud account.

Overall score
--/100
0 of 30 answered

1. Resource hierarchy and organization policy

Guardrails that new projects inherit.

  • Projects live under one organization, grouped into folders by environment, and no production project sits outside the hierarchy.

  • Preventive organization policy constraints are set at the organization or folder level, not project by project.

  • New projects skip the default network (constraints/compute.skipDefaultNetworkCreation) and existing default networks are removed.

  • Resource locations are restricted (constraints/gcp.resourceLocations) wherever data residency matters.

  • Infrastructure changes go through version control, review and a pipeline with policy-as-code checks.

2. Identity and access

Who can do what, and who can be granted access.

  • Users come from a central identity provider with single sign-on, and IAM roles are granted to groups by job function, not to individuals.

  • Domain restricted sharing (constraints/iam.allowedPolicyMemberDomains) limits who can be added to IAM policies, and outside accounts have been reviewed.

  • Basic roles (Owner, Editor, Viewer) are not used for people or workloads in production; narrower roles are used instead.

  • IAM role recommendations are reviewed on a schedule and applied or dismissed with a reason.

  • Humans have view-only access to production; changes go through a pipeline that uses privileged service accounts.

3. Service accounts and workload identity

Machine credentials are the common way in.

  • Service account key creation and upload are disabled by organization policy, with narrow documented exceptions.

  • External and Kubernetes workloads use Workload Identity Federation instead of exported keys.

  • Every external workload identity provider has an attribute condition, and access is not granted to all identities in a pool.

  • Default service accounts do not hold Editor or Owner, and workloads do not share the default Compute Engine service account.

  • You know what to do when a key leaks: delete it quickly, and detect keys unused for 90 days or more.

4. Network and data perimeter

Reduce reachable surface and block exfiltration paths.

  • Projects holding sensitive data in Cloud Storage, BigQuery and similar services sit inside a VPC Service Controls perimeter.

  • Perimeter changes are tested in dry run mode before enforcement, and perimeter edits are monitored.

  • Network resources are centrally managed (for example Shared VPC) and separated by environment.

  • External IPs on VMs are denied by policy (constraints/compute.managed.vmExternalIpAccess) and OS Login is required.

  • No firewall allows SSH or Cloud SQL connections from all addresses; OPEN_SSH_PORT and PUBLIC_SQL_INSTANCE findings are clear.

5. Logging and detection

Evidence you can trust and findings someone sees.

  • Audit logs from every project are routed by organization-level aggregated sinks to a central logging project.

  • Data Access audit logs are enabled for services holding sensitive data.

  • Security Command Center is activated at the organization level and its findings reach a team with owners.

  • Findings are exported to your SIEM or SOC through Pub/Sub or BigQuery.

  • You have rehearsed the response to a leaked key and a compromised workload identity.

6. Data protection and encryption

Storage exposure and key control.

  • Public access prevention (constraints/storage.publicAccessPrevention) is enforced for Cloud Storage.

  • Uniform bucket-level access is required (constraints/storage.uniformBucketLevelAccess) so IAM alone controls access.

  • Where required, CMEK is used with a plan for key disablement, and organization policy controls which services and key projects may be used.

  • Cloud KMS keys are rotated and none is publicly accessible (KMS_KEY_NOT_ROTATED, KMS_PUBLIC_KEY clear).

  • Keys and application secrets live in centralized projects managed and audited by a dedicated team.

Your top gaps

Mark controls Partial or Missing and the highest-weighted gaps appear here, with the reason each matters.

Answers stay in this browser (local storage). Nothing is sent to PlayCISO.

TL;DR

What this is

5 points
  • A self-assessment of 30 Google Cloud controls in 6 domains: resource hierarchy and organization policy, identity and access, service accounts and workload identity, network and data perimeter, logging and detection, and data protection.
  • Mark each control In place, Partial, Missing or N/A. You get a weighted 0 to 100 score overall, a score per domain and your top gaps.
  • Each control carries a short reason and a link to the Google Cloud documentation it comes from. Our own reasoning is labelled PlayCISO practice.
  • Runs in your browser; answers stay in local storage and nothing is sent to PlayCISO. Markdown and PDF export of the report is available with a paid plan.
  • It is not the CIS Google Cloud Platform Foundation Benchmark and not an audit; it connects to nothing in your account.

Frequently asked questions

What does the GCP Security Checklist cover?

It covers 30 controls in 6 domains drawn from Google Cloud documentation: organization policy and resource hierarchy, identity and access, service accounts and Workload Identity Federation, VPC Service Controls and network defaults, audit logging and Security Command Center, and Cloud Storage and Cloud KMS protection.

How is the score calculated?

Each control has a weight of 1 to 3. In place counts as full weight, Partial as half, Missing as zero, and N/A is left out. A domain score is the weighted share of its answered controls and the overall score is the weighted share across all answered controls. Unanswered controls are not scored.

Is this the CIS GCP Foundations Benchmark?

No. The CIS Google Cloud Platform Foundation Benchmark is a separate, versioned document (the CIS page shows version 5.0.0 as the latest at our access date). This checklist is a quick gap review based on Google Cloud documentation. For recommendation numbers and audit evidence, use the benchmark itself and a scanner such as Security Command Center.

Does the tool connect to my Google Cloud account?

No. Nothing connects to Google Cloud and no credentials are requested. You answer from what you know and the score is computed in your browser, so it is safe to use while reviewing a real organization.

Are my answers stored or sent anywhere?

Answers are kept in this browser's local storage and are not sent to PlayCISO. The Markdown and PDF reports are built in your browser.

Why is service account key control weighted so highly?

Google explains that a service account key works like a password and service accounts are not subject to 2-step verification, so a leaked key is likely to succeed. Keys leak through repositories, buckets and file shares, which is why disabling key creation by policy carries the top weight.

Sources

All read on 4 November 2026. Lines in the feedback marked “PlayCISO practice” are our own reasoning, not a Google statement.

The tools people use most on PlayCISO.

All tools
🖤 Black Friday: new customers get 3 months of Pro for $5, then $5/month. Ends 30 Nov. Claim 3 months for $5 →