IAM policies decide what each person or role can do, but an account administrator can rewrite them. If an IAM policy is the only thing keeping your data in Australia or keeping CloudTrail on, any admin in that account can remove it.
Service control policies (SCPs) sit above IAM. You attach them in AWS Organizations to the root, an OU or an account, and they cap what every user and role there can do, including account administrators and the root user. An SCP never grants anything. A request succeeds only when both the SCPs and an IAM policy allow it. In Azure, the closest match is a deny effect in Azure Policy.
Leave the default FullAWSAccess policy attached and add deny statements on top:
organizations:LeaveOrganization and account:CloseAccount. An account that leaves the organization leaves every other guardrail behind. Organizations created in the console after 10 July 2026 get this SCP at the root automatically. Older ones, and ones created with the CLI, an SDK or CloudFormation, do not, so check yoursiam:CreateUser and iam:CreateAccessKey for everyone except your deployment role, so nobody can mint long-lived credentials by hand. People sign in with SSO and workloads use roles. Before you attach this, check for workloads that still need an IAM user, such as Amazon SES SMTP credentialsIAM, Organizations, Route 53, CloudFront, billing and Support are global services. IAM, for example, has a single endpoint in us-east-1, so a plain "deny everything outside Sydney" blocks every IAM call. Use a NotAction list of the global services you use.
This policy restricts Northwind to Sydney and Melbourne, protects CloudTrail and stops accounts leaving or closing themselves. The deployment role is exempt from the region restriction, because some resources must live elsewhere. Root sign-in alerts, for example, only work in three US Regions.
{"Version": "2012-10-17","Statement": [{"Sid": "DenyOutsideAllowedRegions","Effect": "Deny","NotAction": ["iam:*","sts:*","organizations:*","support:*","budgets:*"],"Resource": "*","Condition": {"StringNotEquals": {"aws:RequestedRegion": ["ap-southeast-2", "ap-southeast-4"]},"ArnNotLike": {"aws:PrincipalArn": "arn:aws:iam::*:role/northwind-deploy"}}},{"Sid": "ProtectCloudTrail","Effect": "Deny","Action": ["cloudtrail:StopLogging","cloudtrail:DeleteTrail","cloudtrail:UpdateTrail","cloudtrail:PutEventSelectors"],"Resource": "*","Condition": {"ArnNotLike": {"aws:PrincipalArn": "arn:aws:iam::*:role/northwind-deploy"}}},{"Sid": "DenyLeaveOrCloseAccount","Effect": "Deny","Action": ["organizations:LeaveOrganization","account:CloseAccount"],"Resource": "*"}]}
Warning: The NotAction list above is cut down to 5 entries. Do not deploy it as it is. The AWS region restriction example exempts more than 40 services and actions. Start from that list and remove the services your accounts do not use.
Figure: The region statement in this rule is abbreviated - copy the full exemption list from AWS
For Amazon Bedrock cross-region inference, see Do you keep Amazon Bedrock inference in your sovereignty?
A wrong SCP either blocks production or lets through what you meant to stop. AWS recommends you never attach an SCP to the organization root until you have tested it on a smaller OU.
A Northwind admin edits the region SCP in the console and deletes the iam:* line by mistake. The policy is on the organization root, so IAM calls fail in every account, production included.
❌ Figure: Bad example - Bad example - An untested edit at the root breaks every account at once
The same edit arrives as a pull request. The pipeline attaches it to the NonProd OU, where a test deployment fails on iam:GetRole. The author restores the line before the change reaches Prod.
✅ Figure: Good example - Good example - The mistake is caught in NonProd, where it only breaks a test deployment