RCPs cap what anyone can do to your resources, including callers outside your organization
Facts read 10 Nov 2026. Page reviewed 16 Nov 2026.
AWS added resource control policies to set the maximum permissions on resources in member accounts, whoever the caller is. SCPs limit your own principals; RCPs limit access to your resources, including from outside the organization. This guide covers scope, the supported services, the limits and the data perimeter use.
RCPs limit resources, SCPs limit principals
AWS gives a one-line rule. Use an SCP to limit the permissions of IAM principals inside your member accounts. Use an RCP to restrict principals outside your organization that request access to resources in your member accounts. Both are organization policies and neither grants permissions. SCPs sit on the identity side of a request and RCPs on the resource side, so a request to read an S3 object meets the SCPs of the caller's account, when it is in an organization, and the RCPs of the account that owns the bucket.
An RCP is closest in kind to a resource-based policy, but it is attached to the account, OU or root and sets a ceiling over the resource-based policies below it. The effective permissions are the intersection of the RCPs and SCPs with the identity-based and resource-based policies. See the RCP mechanism page for the library entries. AWS states the job of RCPs in one sentence: central control over the maximum available permissions for resources.
RCPs apply whoever the caller is
AWS documents that RCPs affect the effective permissions of principals reaching a resource in a member account whether or not they belong to the organization, and that includes root users. An RCP on a bucket in Account A applies when a user from outside the organization reads it. It does not apply to resources in the outside account when your users reach those.
That reach is the reason the library flags breakage for partners. Require HTTPS for access to resources denies requests over an unencrypted connection, including from trusted partners. Require TLS 1.2 or later for S3 denies older clients wherever they sit. Test a candidate RCP against the partners and vendors that call your buckets and queues, not only against your own workloads. Their failures show up as Access Denied errors in your CloudTrail logs.
What RCPs cannot restrict, and where they apply
RCPs apply to a subset of AWS services. The AWS list includes Amazon S3, AWS STS, AWS KMS, Amazon SQS, AWS Secrets Manager, Amazon DynamoDB, Amazon ECR, Amazon Cognito and others, and a service missing from the list is outside the policy. A service that is not on the list is not covered, so a control for it needs a resource-based policy or a different mechanism. Check the current list on the AWS page at each review.
AWS also documents that RCPs do not affect resources in the management account, calls made by service-linked roles, AWS managed keys in KMS, or the kms:RetireGrant permission. Like SCPs, they cover member accounts designated as delegated administrators. A control that must cover the management account needs another mechanism. Management account resources are the one gap to design around: protect them with resource-based policies and tight access to the account itself.
The limits are tighter than for SCPs
An RCP can be up to 5,120 characters, half the SCP maximum. Only five can be attached to the root, to an OU or to an account, and the AWS managed RCPFullAWSAccess policy is attached automatically when you enable RCPs. You cannot detach it, and it counts toward the five. These counts are why one policy per control fills the slots quickly.
In practice you get four of your own at each level. The library entries are small single-purpose policies, such as Deny S3 customer-provided encryption keys, that you can merge when you run short. AWS documents that all of these limits are hard limits. Put the rules that never change near the root and the exceptions near the accounts that need them, so inherited policies carry most of the weight without counting against the attachment limit.
RCPs are the enforcement half of a data perimeter
The AWS whitepaper describes a data perimeter as a set of always-on guardrails around identities, resources and networks. The RCP sample README states the aim as ensuring only trusted identities access trusted resources from expected networks, and it carries an identity perimeter RCP for the first objective. Together the two policy types give a perimeter its pair: trusted identities on one side and trusted resources on the other.
The same pairing appears across the library: SCPs constrain your principals, RCPs constrain access to your resources. For rollout, AWS suggests attaching RCPs to individual test accounts first, then moving up through OUs, and reviewing CloudTrail for Access Denied errors. Testing a guardrail first covers why the policy simulator is no help here. Run the first test in a sandbox OU, then move to production OUs while watching for Access Denied errors from callers you did not know about.
Related pages
Sources
- AWS Organizations: Resource control policies, read 10 Nov 2026
- AWS Organizations: Service control policies, read 10 Nov 2026
- AWS Organizations: Quotas and service limits, read 10 Nov 2026
- AWS whitepaper: Building a Data Perimeter on AWS, read 10 Nov 2026
- AWS Organizations: Resource control policy examples, read 10 Nov 2026
Last reviewed 16 Nov 2026.