Skip to content
Cloud Governance Tools

30 questions, answered from the data

Thirty answers for the engineer who writes a guardrail and the lead who approves it: what an SCP can and cannot do, how SCPs, RCPs and permissions boundaries differ, Azure Policy deny against audit, Google Cloud dry run, how to test before applying, and how the platforms are ranked.

See the ranking ยท Read the method

What can a service control policy do?
AWS documents an SCP as an AWS Organizations policy type that sets the maximum available permissions for IAM users and roles in member accounts. It grants nothing; the effective permissions are the intersection of the SCP with identity-based and resource-based policies. It applies to users, roles and the root user in member accounts, and AWS says SCPs affect all users and roles in attached accounts, including the root user.
What can a service control policy not do?
An SCP does not grant permissions, so an allow in an SCP does nothing on its own. AWS says SCPs do not affect users or roles in the management account, and do not affect service-linked roles. An SCP also cannot restrict a resource's access by callers outside your organization; that is the job of a resource control policy. The size limit is 10,240 characters.
What is the difference between an SCP and an RCP?
An SCP sets the maximum permissions for the IAM users and roles in member accounts. An RCP sets the maximum permissions for resources in member accounts, and applies to requests against them whether or not the caller belongs to your organization. Both grant nothing, and neither affects the management account or service-linked roles. An RCP is capped at 5,120 characters and five per root, OU or account.
What is the difference between an SCP and a permissions boundary?
An SCP is attached in AWS Organizations and caps every user and role in the accounts beneath it. A permissions boundary is a managed policy attached to a single IAM user or role, and caps what that entity's identity-based policies can grant. A boundary applies to nothing else, and works inside the account, so the two can be used together.
Does an SCP affect the management account?
No. AWS states that SCPs do not affect users or roles in the management account, and the same holds for RCPs and resources in it. A guardrail written as an SCP therefore does not protect the management account, which is one reason AWS Control Tower's documentation notes that the root user and administrators there can do work that controls would otherwise deny.
How many SCPs can I attach, and how large can they be?
AWS documents a maximum SCP size of 10,240 characters, and at most 10 SCPs attached directly to a root, OU or account, with at least one always attached. RCPs are capped at 5,120 characters and 5 per entity, counting the default RCPFullAWSAccess policy. Inside an AWS Control Tower landing zone the limit of 10 SCPs per OU also applies, so combine statements deliberately.
What is an AWS declarative policy?
A declarative policy declares and enforces a baseline configuration for an AWS service in the service's control plane, rather than regulating API access. It also covers actions by service-linked roles, which SCPs do not. For an EC2 declarative policy, account administrators cannot change the attribute and a noncompliant action fails. AWS says detaching the policy rolls the attribute back. The maximum size is 10,000 characters.
Can I test an SCP before attaching it?
Partly. AWS documents that the IAM policy simulator evaluates SCPs, including condition keys and resource scoping in deny statements, for identity-based policies, but it does not surface the matched SCP statements. AWS recommends testing an SCP on a small set of accounts in a separate OU before attaching it near the organization root. Every SCP entry in the library repeats this test step.
Can I test a resource control policy before attaching it?
Not in the IAM policy simulator, which AWS documents as not supporting RCPs. AWS suggests reviewing CloudTrail logs for Access Denied errors before attaching one. Because an RCP applies to callers outside your organization too, the review should include partners and third-party integrations. IAM Access Analyzer reads RCPs in its analysis, and picks up a change at its next periodic scan within 24 hours.
What is the difference between Azure Policy deny and audit?
Deny stops a matching create or update request before it reaches the resource provider. Audit lets the request through and creates a warning event. Microsoft documents that deny does not touch existing resources, which are only marked noncompliant. Modify and deployIfNotExists are different again: they change or deploy settings, and existing resources change only through a remediation task. Assigning with audit first is the lower-risk start.
How do I test an Azure Policy before enforcing it?
Assign the definition with enforcementMode set to DoNotEnforce, which Microsoft documents as the way to evaluate an assignment before enforcement, or assign it with the audit effect first. Review the compliance results, fix or exempt what shows up, then switch the mode. Remember that effects are evaluated in order, so an earlier modify or append can change the request before a deny is considered.
Can I create Azure deny assignments myself?
Microsoft documents that customer-managed deny assignments are in private preview, and that for generally available scenarios Azure creates and manages them, for example when you set deny settings on a deployment stack. A stack can exclude at most five principals and its deny settings cover only resources the template defines explicitly. No dry run was found for deny assignments on 2026-11-10.
Should I use an Azure resource lock or a policy?
They solve different problems. A lock blocks deletion (CanNotDelete) or modification and deletion (ReadOnly) of a scope, and overrides user permissions, but only for control plane operations. A ReadOnly lock can block a virtual machine from starting, because that uses POST. A policy evaluates the content of a request. Locks cannot be added to management groups, and Owners can remove them.
How does Google Cloud organization policy dry run work?
Google documents organization policies in dry-run mode: violations are audit logged but not denied, so you can see what a constraint would block before enforcing it. Dry run covers custom constraints, managed constraints and a short list of legacy managed constraints. Policy Simulator for Organization Policy also lists the resources that would violate a proposed policy, but it cannot test changes to legacy managed constraints.
Are Google Cloud organization policies retroactive?
Usually not. Google says new organization policies are usually not retroactive, so existing violations are not stopped, and the library notes cases such as Cloud SQL public IP access and Vertex AI Workbench settings, which apply only to new instances. Cloud Storage public access prevention is the exception in the library: Google documents that it revokes public access for existing buckets and objects once enabled.
In what order does Google Cloud check deny and allow policies?
Google documents that IAM checks relevant deny policies before allow policies, so a deny blocks a permission regardless of the roles granted. Principal access boundary policies work the same way: they define the resources a principal is eligible to access and can only block. Policy Troubleshooter examines the allow, deny and principal access boundary policies together and says whether a principal can use a permission.
What does VPC Service Controls not block?
Google documents that a service perimeter blocks communication to Google Cloud managed services across the boundary, but not third-party APIs or services on the internet, and that not all Google Cloud services are supported. Changes can take up to 30 minutes to propagate and may briefly block requests. Dry run mode logs requests that would violate the perimeter without denying them.
What is a custom organization policy constraint?
It is a constraint you define in a YAML file that names the resources, methods, conditions and actions, with the condition written in Common Expression Language (CEL). You then enforce it with an organization policy. Most resource types support up to 20 custom constraints, a condition holds up to 1,500 characters, and a policy set with set-policy can take up to 15 minutes to take effect.
What is the difference between a preventive and a detective control?
A preventive control answers a request with a denial before it takes effect: an SCP, an Azure Policy deny, a Google Cloud constraint. A detective control reports after the fact. AWS Control Tower shows the split: preventive controls use SCPs and RCPs, detective controls use AWS Config, and proactive controls use CloudFormation hooks, which only cover CloudFormation stack operations.
How do I test a guardrail before applying it?
Use the mechanism's own test first: the IAM policy simulator and a separate test OU in AWS, enforcementMode DoNotEnforce or the audit effect in Azure, dry-run mode and Policy Simulator in Google Cloud. Then read the log for the period before you enforce. The Simulation Coverage Matrix shows which platforms add a test of their own, and each library entry names the test step and its source.
Where do the 100 guardrails come from?
From provider documentation and from open source sets with permissive licenses. Sixty are taken from the aws-samples service control policy and resource control policy repositories (MIT-0) and the Azure/azure-policy repository (MIT License), and 40 are written from provider documentation. Nothing is invented, each entry links to its file or page, and the dates read are 10 Nov 2026.
Can I copy the policies in the library?
The aws-samples repositories state MIT-0, and the LICENSE file read is MIT-0 text, copyright Amazon.com, Inc. The Azure/azure-policy repository is under the MIT License. Entries written from provider documentation are examples of what the documentation describes. Each entry states its license, and every policy needs testing in your environment first, since the placeholders in many must be replaced.
How are the platforms ranked here?
On seven criteria with weights fixed before any score existed: safe to enforce 30, one change for a class of risk 30, removes access 15, stays fixed 10, uses the cloud's own controls 10, evidence for auditors 3, and identity, network and AI coverage 2. The total is a weighted mean rounded to one decimal. Act leads at 5.9, then Sonrai Security at 5.7 and AWS Control Tower at 4.2.
Why does Act rank first?
Act scores 7 on safe to enforce, the only 7 and the criterion with the largest weight. Act says each change is tested against a digital twin built from cloud configuration data, then run in audit mode against audit-log events. It also scores 7 on one change for a class of risk, with guardrail campaigns at the SCP level. Act is scored by the same rules as every other platform.
Where is Act behind?
Act writes guardrails in the provider's own mechanisms and hands them to your pipeline to enforce, so it is bounded by what those mechanisms can express in each cloud. It does not apply them itself, the boundary types it names are AWS's and Azure's, and exception and waiver handling after a guardrail ships was described as roadmap. On removes, Act scores 3 against 5 for Sonrai.
Why do Sonrai and Act tie on regulated and small-team?
Both total 5.2 on the regulated view and 5.5 on the small-team view at one decimal, so they share first place. Sonrai is listed first because it is ahead before rounding: its removes lead, and its edge on native controls and audit evidence, slightly exceed Act's lead on safe to enforce and on identity, network and AI coverage. Sonrai's pages say it deploys the blocking policy and updates it on approval; Act delivers a pull request for your team to apply.
Why is AWS Control Tower third?
Control Tower scores 4.2. It applies preventive controls to whole OUs from a catalog of more than 750 controls through SCPs and RCPs, and scans for drift daily. It is AWS only, has a management-account bypass, is limited to 10 SCPs per OU, and AWS tells users to test proactive controls themselves, with no impact simulation described. That is why it scores 1 on safe to enforce.
Does any platform simulate a guardrail against my real environment?
Act and Google Cloud's Policy Simulator are the clearest cases in the evidence. Act says it tests changes on a digital twin of your configuration and in audit mode on your audit logs. Google lists resources that would violate a proposed organization policy before enforcement, for Google Cloud only. CrowdStrike states a dry run without saying what it runs against. Sonrai shows its exemption list before deployment but is not described as simulating.
Do the just-in-time access tools write guardrails?
Not in the evidence read. Apono can place a dormant principal in an AWS permission boundary and gives a deny rule for the user to create. Britive scopes access by profile; a deny or boundary: none found on 2026-11-10. P0 has a deny-all policy that stops requests through P0 and not a cloud boundary. Entitle sets approval workflow, owner and requestable status. They are ranked here on that basis.
What happened to Microsoft's CIEM product?
Microsoft is scored on cloud infrastructure entitlement management in Defender for Cloud, because Entra Permissions Management was retired on 2025-11-01 and has not been sold since 2025-04-01. Defender for Cloud gives recommendations described as guidance for remediation. Azure Policy is a separate Azure service, and Microsoft's pages do not describe it acting on CIEM findings.

Last reviewed 16 Nov 2026.