Most Google Cloud security problems trace back to a foundation that was never designed: projects created by hand, the default network left in place, owners granted to individuals and logs kept wherever they landed. Google's enterprise foundations blueprint (formerly the security foundations blueprint) is the provider's reference answer: an architecture guide plus a deployable Terraform repository, terraform-example-foundation, that builds the organisation hierarchy, identity model, networks, preventive policies and detective controls before any workload arrives. This guide explains what the blueprint actually deploys, where its security value lies, the decisions you must make before running it, and the places teams most often weaken it. It is for platform engineering, cloud security architects and auditors adopting or reviewing a Google Cloud foundation. For the provider-neutral design principles, see the Cloud Landing Zone Security Guide; details here were checked against Google's documentation (last updated 15 May 2025) and repository release 7.0.0 (6 October 2026) on 27 October 2026.
This is an independent guide from NHI Mgmt Group, based on Google Cloud's public documentation as of November 2026. It is not sponsored or reviewed by Google Cloud. Product features change, so confirm settings against Google Cloud's current documentation.
Key takeaways
- It is a starting point, not a product. Google positions the blueprint as an opinionated example to customise; you own every decision once it is deployed.
- The pipeline is the control plane. After a manual bootstrap, every change flows through a foundation pipeline with policy checks, so the pipeline's service accounts are the most privileged identities in the organisation.
- Preventive controls are strong by default. Nineteen organisation policy constraints at the organisation node block key creation, public buckets, external VM IPs, public Cloud SQL and default network creation.
- Two important controls ship switched off. VPC Service Controls is deployed in dry run, and Data Access audit logs are off; both need deliberate follow-up.
- Exceptions decide its real strength. The blueprint expects narrow, approved organisation policy exceptions; a loose exception process undoes the baseline quickly.
What the blueprint is and how it is built
The blueprint has two parts: an architecture guide covering authentication and authorisation, organisation structure, networking, detective controls, preventive controls, deployment methodology and operations; and the Terraform repository built on Cloud Foundation Toolkit modules. Google describes the defence-in-depth model as three layers: policy controls (infrastructure-as-code validation and organisation policy constraints), architecture controls (resource hierarchy and network design) and detective controls (Security Command Center and custom detection).
Deployment stages
| Stage | What it creates | Security significance |
|---|---|---|
0-bootstrap | Folder fldr-bootstrap; seed project prj-b-seed (Terraform state bucket and stage service accounts) and CI/CD project prj-b-cicd | Run manually once; creates the identities that will hold organisation-wide power |
1-org | fldr-common with prj-c-logging, prj-c-billing-export, prj-c-scc, prj-c-kms, prj-c-secrets; fldr-network with DNS, interconnect and Shared VPC host projects; organisation policies; VPC Service Controls (dry run by default) | Sets the baseline every project inherits |
2-environments | Development, non-production and production folders, each with its own KMS and secrets projects | Environment separation at the hierarchy level |
3-networks-svpc or 3-networks-hub-and-spoke | Shared VPC per environment with firewall baseline, Private Service Connect to Google APIs, no default internet route, optional Cloud NAT, DNS hub | Removes direct internet paths and prepares for VPC Service Controls |
4-projects | Business unit service projects attached to Shared VPC, plus per-unit infrastructure pipeline projects | Where workload teams get their landing space |
5-app-infra (optional) | A sample Compute Engine instance deployed through the infrastructure pipeline | Demonstrates the workload deployment path |
Naming conventions are part of the design: folders fldr-environment, projects prj-environmentcode-description-randomid, groups grp-gcp-description@, custom roles rl-description and service accounts sa-description@. Consistent names make organisation policy exceptions, log queries and access reviews far easier.
Identity, groups and privileged service accounts
The blueprint assumes users live in an existing identity provider, for example Active Directory synchronised to Cloud Identity with Google Cloud Directory Sync, with SAML single sign-on and no password synchronisation. Access is granted to groups, never individuals.
| Group | Role and scope | Use |
|---|---|---|
grp-gcp-org-admin | Organization Administrator, organisation | Granting IAM at organisation level; not for daily use |
grp-gcp-billing-admin | Billing Account Administrator | Billing changes; not for daily use |
grp-gcp-billing-viewer | Billing Account Viewer; BigQuery User on billing project | Cost analysis |
grp-gcp-security-reviewer | Security Reviewer, organisation | Security review of IAM and configuration |
grp-gcp-network-viewer | Compute Network Viewer, organisation | Network operations visibility |
grp-gcp-audit-viewer | Logs Viewer and BigQuery User, logging project | Audit log review |
grp-gcp-scc-admin | Security Center Admin Editor, organisation | Security Command Center configuration |
grp-gcp-secrets-admin, grp-gcp-kms-admin | Secret Manager Admin on secrets projects; Cloud KMS Viewer on KMS projects | Secrets and key governance |
In the Terraform configuration, the organisation admin, billing admin, billing data and audit data groups are required and the others optional; the repository can also accept workforce identity federation principal sets in place of Google Groups.
The pipeline identities
- Stage service accounts:
0-bootstrapcreates one service account per stage in the seed project (sa-terraform-bootstrap,-org,-env,-net,-proj). Between them they can change the organisation, folders, networks and every project. Treat them as tier-zero NHIs. See the CI/CD Pipeline Identity Security Guide. - No keys: the pipeline runs as these accounts through Cloud Build, or through workload identity federation when you use GitHub Actions or Terraform Cloud; operators use impersonation for troubleshooting. The organisation policy
iam.disableServiceAccountKeyCreationkeeps it that way. See the Workload Identity Federation Guide. - Restrict impersonation: limit who holds Service Account Token Creator on the stage accounts to the pipeline and a break-glass group, and alert on any other use.
- Super admins: Cloud Identity super admins bypass SSO by design so they can recover from an SSO outage. Google's guidance is to enforce 2-step verification with security keys, disable super admin account recovery and keep an internal recovery process. See the Break-Glass Account Guide.
Preventive controls: organisation policies and policy as code
The blueprint applies nineteen organisation policy constraints at the organisation node so every folder and project inherits them. The most security-relevant are below; for how constraints and exceptions work, see the Google Cloud Organization Policy Guide.
| Constraint | Effect | Risk it reduces |
|---|---|---|
iam.disableServiceAccountKeyCreation, iam.disableServiceAccountKeyUpload | No user-managed service account keys | Leaked long-lived machine credentials |
iam.allowedPolicyMemberDomains | Allow policies limited to your managed domain | Data shared with personal or external accounts |
iam.automaticIamGrantsForDefaultServiceAccounts | No automatic Editor grant to default service accounts | Over-privileged workloads by default |
storage.publicAccessPrevention, storage.uniformBucketLevelAccess | No allUsers or allAuthenticatedUsers grants; IAM only, no legacy ACLs | Public bucket exposure |
compute.vmExternalIpAccess (empty allowlist), compute.skipDefaultNetworkCreation | No external IPv4 on VMs; no default network with permissive rules | Directly exposed instances |
compute.requireOsLogin, compute.disableSerialPortAccess | IAM-controlled SSH; no serial console | Unmanaged SSH keys and side-channel access |
sql.restrictPublicIp, sql.restrictAuthorizedNetworks | No public IP; Cloud SQL Auth Proxy instead of authorised networks | Internet-exposed databases |
essentialcontacts.allowedContactDomains | Security notifications only to approved domains | Google security notices sent to former staff or outsiders |
- Policy as code in the pipeline: Open Policy Agent constraints, forked from
GoogleCloudPlatform/policy-library, are evaluated by Google Cloud CLI policy validation duringcloudbuild-tf-plan.yaml. If the plan fails, the apply build deploys nothing. - Optional controls for sensitive workloads: resource location restrictions, Assured Workloads, Key Access Justifications and disabling Cloud Shell. See the Assured Workloads and Sovereign Controls Guide.
- VPC Service Controls: configured by
1-orgin dry run by default because enforcement can be disruptive. When violations are resolved, setenforce_vpcsctotruein theservice_controlmodule and re-apply1-org, then2-environmentsand4-projects. See the VPC Service Controls Guide.
Network architecture
| Option | How it works | Choose it when |
|---|---|---|
Shared VPC per environment (3-networks-svpc) | One Shared VPC per environment, subnets in two regions, no direct traffic between environments | Environments should be isolated and cross-environment flows are rare; simpler to operate |
Hub-and-spoke (3-networks-hub-and-spoke) | Hub connected to spokes by VPC Network Peering; network virtual appliances behind internal load balancers inspect spoke-to-spoke traffic; on-premises only via the hub | You need central inspection or third-party firewalls, and can run the appliances |
- No default internet: no direct inbound or outbound internet by default. Outbound uses Cloud NAT or Secure Web Proxy for allowlists; inbound uses Cloud Load Balancing with Google Cloud Armor.
- Private API access: a Private Service Connect endpoint using the
vpc-scAPI bundle, with private DNS zones forgoogleapis.com,gcr.ioandpkg.dev, and Private Google Access on every subnet. This is a prerequisite for VPC Service Controls. - Firewall policy hierarchy: a hierarchical firewall policy on the environment, bootstrap and common folders allows IAP TCP forwarding from
35.235.240.0/20on TCP 22 and 3389 and load balancer health checks, and delegates RFC 1918 decisions downward. Each VPC's network firewall policy allows egress only to the PSC endpoint on TCP 443 and denies the rest. Tags with IAM controls replace legacy network tags. See the Google Cloud VPC and Cloud NGFW Security Guide. - Hybrid: a DNS hub forwards between Cloud DNS and on-premises DNS; the reference connectivity is Dedicated Interconnect in a 99.99 percent availability pattern, with MACsec or HA VPN over Interconnect for encryption.
Detective controls and logging
- Central logging: organisation-level aggregated sinks send logs to
prj-c-logging:sk-c-logging-la(a Log Analytics bucket with a linked BigQuery dataset),sk-c-logging-bkt(Cloud Storage for long-term retention) andsk-c-logging-pub(Pub/Sub to an external SIEM). See the Google Cloud Audit Logs and Log Sinks Guide. - Data Access logs are off: the blueprint does not enable them by default; enable them for services handling sensitive data after deployment. VPC Flow Logs are sampled at 50 percent.
- Security Command Center: activated after deployment (Google suggests Premium), with a Pub/Sub topic in
prj-c-sccfor SIEM or SOAR export. See the Security Command Center Guide. - IAM change monitoring: a Cloud Asset Inventory feed triggers a Cloud Run function that raises a custom finding when Organization Admin, Owner or Editor is granted, with an optional automatic revert.
- Log-based custom findings: the
bq-log-alertingsample runs scheduled BigQuery queries every 15 minutes and writes custom findings to Security Command Center. - Cost as a signal: budgets alert when forecast spend reaches 120 percent; Google's operations guidance treats unexpected spend as a possible security incident.
Metrics
- Percentage of foundation changes deployed through the pipeline versus manual changes found in Admin Activity logs (target 100 percent pipeline).
- Number of active organisation policy exceptions, each with owner, reason and review date.
- Projects in an enforced VPC Service Controls perimeter versus dry run.
- Projects with sensitive data that have Data Access logs enabled.
- Principals other than the pipeline with Token Creator on stage service accounts (target zero).
Rolling out and operating the foundation
- Decide before you deploy: identity source and group names, network option, regions, CI/CD tool (Cloud Build by default; GitHub Actions and Terraform Cloud documented; the repository states GitLab guidance is no longer actively tested), and which optional controls apply.
- Bootstrap in a clean organisation or a new folder: run
0-bootstrapmanually with a temporary highly privileged user, then remove that user's standing access once the pipeline works. - Promote through branches: three persistent branches (development, non-production, production) mirror the environments; changes move by approved pull request, with an optional manual approval step in Cloud Build before apply.
- Test controls in the lower environment first: validate organisation policies and firewall policy in development, and run VPC Service Controls in dry run while collecting violations.
- Complete post-deployment steps: activate Security Command Center, enable Data Access logs where needed, and consider Access Transparency and Access Approval.
- Roll back through code: revert the pull request and let the pipeline re-apply. For an emergency, use pre-planned break-glass access for super admin, organisation admin, pipeline admin and SRE roles, with approvals, alerting (for example a custom Event Threat Detection module) and revocation afterwards.
- Keep current: track repository releases (major versions arrive regularly, 6.0.0 and 7.0.0 were a month apart in 2026) and review the changelog before upgrading.
Small versus large organisations: a small team can deploy the Shared VPC option with a single infrastructure pipeline and Security Command Center Standard, while a large enterprise typically adopts hub-and-spoke, one infrastructure pipeline per business unit, Premium detection and scoped VPC Service Controls perimeters.
Common failure patterns
- Dry run forever:
enforce_vpcscis never set totrue, so the exfiltration control exists only on paper. - Bootstrap identity left behind: the user who ran
0-bootstrapkeeps Organization Administrator or Owner on the seed project, giving an attacker a shortcut around the pipeline. - Exceptions by default: teams blocked by
iam.disableServiceAccountKeyCreationorcompute.vmExternalIpAccessget folder-wide exceptions instead of narrow, project-level ones with compensating controls. - Manual drift: console changes made during incidents are never written back to code, and the next pipeline run either reverts a fix or the code no longer reflects reality.
- Pipeline credentials exposed: when the foundation runs on an external CI system, tokens leak through build output. Use federation, never keys, and keep artifacts free of credentials.
- Logging assumed complete: teams assume the central sinks capture data access, but Data Access logs were never enabled.
Framework mapping
| Framework | Reference | Blueprint element |
|---|---|---|
| NIST SP 800-53 Rev. 5 | CM-2 Baseline Configuration; CM-3 Configuration Change Control; PL-8 Security and Privacy Architectures | Terraform baseline, pipeline promotion, reference architecture |
| NIST SP 800-53 Rev. 5 | AC-2 Account Management; AC-6 Least Privilege; AU-9 Protection of Audit Information | Group-based IAM, per-stage service accounts, central logging project |
| NIST CSF 2.0 | PR.PS-01; PR.AA-01; PR.AA-05; PR.IR-01; DE.CM-09 | Configuration management, identity management, network protection, monitoring |
| CIS Controls v8.1 | 4.1 Secure Configuration Process; 5.4 Restrict Administrator Privileges to Dedicated Administrator Accounts; 8.9 Centralize Audit Logs | Organisation policies, admin groups, aggregated sinks |
| ISO/IEC 27001:2022 Annex A | A.5.15 Access control; A.8.9 Configuration management; A.8.32 Change management | Groups, policy constraints, pull-request promotion |
| MITRE ATT&CK | T1078.004 Cloud Accounts; T1098.003 Additional Cloud Roles | Keyless pipeline identities; IAM grant monitoring with revert |
Practitioner checklist
- Record every deviation from the blueprint with an owner and reason before deployment.
- Confirm all nineteen organisation policy constraints are enforced at the organisation node, and list every exception.
- Remove standing human access to the seed and CI/CD projects after bootstrap.
- Restrict Token Creator on the
sa-terraform-*accounts to the pipeline and break-glass, and alert on other use. - Plan the VPC Service Controls enforcement date and set
enforce_vpcscwhen violations are cleared. - Enable Data Access logs for sensitive services and confirm all three aggregated sinks receive logs.
- Activate Security Command Center and route the
prj-c-scctopic to your SOC. - Protect super admins with security keys and disabled account recovery.
- Require approved pull requests on all persistent branches and review drift monthly.
- Track repository releases and test upgrades in development first.
Standards and references
- Google Cloud: Enterprise foundations blueprint
- Google: terraform-example-foundation repository
- Google Cloud: Enterprise foundations blueprint, authentication and authorisation
- Google Cloud: Enterprise foundations blueprint, preventive controls
- Google Cloud: Enterprise foundations blueprint, detective controls
- Google Cloud: Enterprise foundations blueprint, networking
- Google Cloud: Enterprise foundations blueprint, operations best practices
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: Cloud Landing Zone Security Guide · Google Cloud Security Guide · Google Cloud IAM and Service Accounts Guide · CIS Cloud Foundations Benchmarks Guide