# Google Cloud splits guardrails into four layers: organization policy for what, IAM deny for who, and perimeters for data

Facts read 10 Nov 2026. Page reviewed 16 Nov 2026.

Google Cloud organization policy restricts how resources can be configured through constraints set on an organization, folder or project. This guide covers constraint types, inheritance and merging, dry-run mode, custom constraints, and how IAM deny policies and VPC Service Controls cover the cases that organization policy does not.

## Organization policy controls what, IAM controls who

Google describes the split plainly: IAM focuses on who, and lets an administrator say which principals can act on which resources. Organization Policy focuses on what, and sets restrictions on how resources can be configured. A policy sets one constraint on an organization, folder or project and its descendants inherit it. Both are inherited down the hierarchy, so a setting on the organization reaches every project.

A constraint is the blueprint of a controlled behavior, and the organization policy is the rule set that enforces it. The library groups the entries under [the organization policy mechanism](/mechanisms/org-policy/). A policy can hold several rules, for example one that enforces a constraint only on resources tagged environment=development. The constraint list is long, and Google frames it as fine-grained control over specific resource behaviors.

## Constraints come as managed, legacy managed and custom

Managed constraints are newer constraints with list or boolean parameters. Legacy managed constraints use list rules, a collection of allowed or denied values, or boolean rules, which can allow all values, deny all values or check enforcement. Google notes managed constraints are designed to replace equivalent legacy ones. Managed constraints support simulation in Policy Simulator by default, and the documentation notes any that do not.

Custom constraints are defined by your organization in a YAML file that names the resources, methods, conditions and actions. A custom constraint ID must start with custom. and cannot change after creation. [The custom organization policy mechanism](/mechanisms/custom-org-policy/) holds the library example that limits IAM role grants to organization members. Custom constraints exist only for the services Google lists as supporting them, so check the supported-service list before you plan one.

## Inheritance merges or replaces, and deny wins

By default descendants inherit the policy of an ancestor. A child can set inheritFromParent to true, in which case the parent policy is merged with its own, or to false, in which case it ignores the parent and uses the constraint default unless it sets its own values. Google states that the default behavior is never merged. Read the effective policy at the project, not the policy at the organization.

When values are merged and conflict, the DENY value takes precedence. Managed constraints inherited from a parent are not merged: the resource enforces the constraint as set on the parent or overrides it with its own policy. If a constraint has a Google-managed default that restricts an operation, the operation is restricted even if you never wrote a policy for it.

## Dry-run mode logs violations without denying them

An organization policy in dry-run mode is created and enforced like any other, but violations are audit logged and the violating actions are not denied. You need billing enabled on the project, the Organization policy administrator role and the Logs Configuration Writer role. Only custom constraints, managed constraints and certain legacy constraints support dry-run. Dry-run rules are set apart from the live policy on the constraint.

Policy Simulator for Organization Policy goes one step earlier. It previews a new custom or managed constraint and lists the resources that violate the proposed policy before it is enforced, so you can reconfigure resources, request exceptions or narrow the scope. Run it first, then dry-run, then enforce. If a constraint is not available in dry-run, the simulator or a lower environment is the only preview.

## IAM deny and VPC Service Controls cover what constraints do not

IAM deny policies attach to a project, folder or organization and hold deny rules for principals and permissions. Google documents that IAM always checks relevant deny policies before allow policies, so a denied permission cannot be restored by granting a role. The library has [Deny deleting projects tagged prod](/guardrails/gcp-deny-project-delete-prod/) as an example. Deny policies are eventually consistent, so allow time for a change to propagate.

VPC Service Controls is a separate layer around services: a service perimeter controls data egress across a boundary, independent of IAM, and Google recommends using both. It has its own dry run mode in which requests that violate the perimeter are logged and not denied. See [Service perimeter around managed services](/guardrails/gcp-vpcsc-service-perimeter/). The library lists both under their own mechanism pages.

## Principal access boundaries limit the resources a principal can reach

Google documents another control that sits beside deny: principal access boundary policies. By default, principals are eligible to access any Google Cloud resource. A Principal Access Boundary policy defines the resources a principal is eligible to access, and if a principal is ineligible, access is limited regardless of the roles it holds. Google lists this as part of IAM, not of organization policy, which is why it appears here as a separate layer.

These policies attach to principal sets, unlike allow policies, which attach to resources. Google gives preventing principals from reaching resources in other organizations as an example, which can help prevent phishing. See [Principal access boundary limiting the resources principals can access](/guardrails/gcp-pab-eligible-resources/) and [the mechanism page](/mechanisms/principal-access-boundary/). Check whether the principal sets you need are supported before you design around it. Both are small checks.

## Related pages

- [Preventive vs detective controls: when to block and when to report](/guides/preventive-vs-detective-controls/)
- [IAM Access Analyzer as a guardrail tool: before and after a change](/guides/iam-access-analyzer-as-a-guardrail-tool/)
- [Testing a guardrail before you apply it: simulators and dry-run modes](/guides/testing-a-guardrail-before-you-apply-it/)
- [All of them](/guides/)

## Sources

- Google Cloud: Introduction to the Organization Policy Service: https://cloud.google.com/resource-manager/docs/organization-policy/overview (read 10 Nov 2026)
- Google Cloud: Understanding hierarchy evaluation: https://cloud.google.com/resource-manager/docs/organization-policy/understanding-hierarchy (read 10 Nov 2026)
- Google Cloud: Test organization policies in dry-run mode: https://cloud.google.com/resource-manager/docs/organization-policy/dry-run-policy (read 10 Nov 2026)
- Google Cloud: Test organization policy changes with Policy Simulator: https://cloud.google.com/policy-intelligence/docs/test-organization-policies (read 10 Nov 2026)
- Google Cloud: Create custom constraints: https://cloud.google.com/resource-manager/docs/organization-policy/creating-managing-custom-constraints (read 10 Nov 2026)
- Google Cloud: IAM deny policies: https://cloud.google.com/iam/docs/deny-overview (read 10 Nov 2026)
- Google Cloud: VPC Service Controls overview: https://cloud.google.com/vpc-service-controls/docs/overview (read 10 Nov 2026)
- Google Cloud: Dry run mode for service perimeters: https://cloud.google.com/vpc-service-controls/docs/dry-run-mode (read 10 Nov 2026)

Last reviewed 16 Nov 2026. Canonical: https://cloudgovernancetools.com/guides/google-cloud-organization-policy/
