AWS IAM Identity Center is the AWS service that gives your people single sign-on to every AWS account and application in your organization, with short-lived credentials instead of IAM users. It comes at no additional charge. If your team still creates IAM users in each account, you're carrying long-term access keys, duplicate user lists and audit trails split across accounts.
Identity Center removes all three, and 2026 removed one of its oldest objections: an instance can now replicate to additional Regions. You'll see when IAM users still make sense, how to retire the rest, and how to plan Regions and quotas. I treat Identity Center as the access layer of any multi-account landing zone, so the advice leans that way.
What is AWS IAM Identity Center?
AWS IAM Identity Center is AWS's workforce identity service, formerly AWS Single Sign-On (AWS SSO), which is why aws sso login still says "SSO". You connect one identity source, define permission sets, and assign users and groups to AWS accounts and applications. With AWS Identity Center, users sign in once at the AWS access portal and pick the account and role they need.
It works with AWS Organizations to reach every member account, and AWS managed applications use it for workforce identity. The service itself has no additional charge; you pay standard AWS KMS rates only if you use a customer managed key, which multi-Region replication requires.
What IdP and SCIM mean
An identity provider (IdP) is the system where your users' identities and passwords already live, such as Okta, Microsoft Entra ID or Google Workspace. Identity Center trusts the IdP over SAML 2.0, so people sign in with their existing corporate credentials.
SCIM (System for Cross-domain Identity Management) v2.0 is the protocol the IdP uses to push users and groups into Identity Center automatically, through a SCIM endpoint and a bearer token you create in Identity Center. SAML handles sign-in; SCIM handles who exists. With an external IdP you almost always want both.
Where those identities land depends on the type of instance you enable.
Organization instances vs account instances
An organization instance is enabled in the AWS Organizations management account. There's one per organization, and it supports every feature. Any instance enabled before November 15, 2023 is an organization instance.
An account instance is bound to a single AWS account and Region. You create it from a member account or a standalone account, and it comes with real restrictions:
- No permission sets and no access to AWS accounts
- No SAML 2.0 customer managed applications (OIDC only), and only select AWS managed applications
- No delegated administrator, no customer managed KMS key and no multi-Region replication
- No conversion to, or merge into, an organization instance, and no moving it out of its account
AWS suggests account instances for temporary trials of a managed application, isolated user populations, or organizations you don't control. For workforce access to AWS accounts, you need an organization instance.
IAM users vs IAM Identity Center: which should you use?
Use IAM Identity Center for people and IAM roles for workloads. Keep IAM users only for the narrow cases where AWS says long-term credentials are unavoidable. That rule holds for any team following a multi-account strategy, and it's easiest to adopt while moving from a single account to multi-account.
| Aspect | IAM Identity Center | IAM users |
|---|---|---|
| Credentials | Temporary role credentials, 1 to 12 hours | Long-term passwords and access keys |
| Multi-account | One identity, assigned to many accounts | A separate user in every account |
| Identity source | External IdP, Active Directory or built-in directory | Standalone in each account |
| MFA | Managed in Identity Center or your IdP | Configured per user, per account |
| Audit | Activity traces back to one named person | Separate users per account to correlate |
| User experience | One portal for every account and application | A console sign-in per account |
| Where it fits | People | The edge cases AWS lists below |
Temporary credentials remove a familiar finding in AWS security audits: years-old access keys in an engineer's credentials file. Keys in CI variables go the same way through OIDC roles, covered in the migration steps. Choose Identity Center when people use more than one AWS account, you run a corporate IdP, AWS managed applications need a workforce identity, or you want long-term keys gone.
AWS's IAM best-practices page gives four example cases that still need IAM users with long-term credentials:
- Workloads that can't use IAM roles (AWS's example is WordPress plugins)
- Third-party clients not hosted on AWS that don't support Identity Center
- AWS CodeCommit SSH keys or service-specific credentials, although the
git-remote-codecommitutility avoids the IAM user - Amazon Keyspaces service-specific credentials where Identity Center users can't be used
If you keep any of these, require phishing-resistant MFA wherever a person signs in, rotate access keys, and use last-used data to retire the keys nobody touches.
Workloads aren't IAM users either: they use IAM roles, or IAM Roles Anywhere and OIDC federation outside AWS.
How IAM Identity Center works
Identity Center sits between your identity source and your AWS accounts. It stores permission sets and assignments, provisions an IAM role for each permission set in every account it's assigned to, and shows each user only what they can reach.
A user opens the access portal and authenticates against the identity source, with MFA where it's configured. The portal lists the accounts and permission sets assigned to them or their groups. They pick one, open the console or copy short-term credentials, and work through the AWSReservedSSO_* role in that account until the permission set's session duration runs out. No step hands out a long-term credential.
Identity sources: directory, Active Directory or external IdP
Your identity source decides where users live and who authenticates them. An instance uses one identity source at a time, so choose based on where your people already are.
Identity Center directory
The built-in directory is the default. You create users and groups in the console, and MFA is on by default for them. It suits teams without a corporate IdP, small teams and proofs of concept. Since July 29, 2026 it also supports multi-Region replication.
Active Directory
You can connect AWS Managed Microsoft AD or a self-managed Active Directory through AWS Directory Service. Two constraints shape the design: the directory's home Region must match the Identity Center Region, and you can connect only one directory, a quota AWS won't raise.
Active Directory is also the one identity source that doesn't support multi-Region replication. If resilience across Regions matters, that's a reason to plan a move to an external IdP. Since October 20, 2026, organization instances can change the identity source away from Active Directory and keep users, groups, memberships and permission set and application assignments.
External IdP with SCIM provisioning
For Okta, Microsoft Entra ID, Ping, JumpCloud or Google Workspace, connect the IdP over SAML 2.0 for sign-in and turn on SCIM so users and groups arrive automatically. MFA then belongs to the IdP.
These SCIM provisioning details are the ones that break synchronization:
- SCIM access tokens are valid for one year. Reminders start 90 days out in the console and the AWS Health Dashboard, and synchronization stops if the token expires. An instance holds up to two tokens, so create the new one before you delete the old.
- Every user needs a first name, last name, username and display name, or they aren't provisioned. Multivalue attributes, such as several emails or phone numbers, fail.
- Users who aren't assigned to the AWS application in your IdP never sync.
- Once SCIM is on, you can't add or edit users in the Identity Center console. Make changes in the IdP.
- Some IdPs, Google Workspace among them, don't provision groups over SCIM, so you manage those groups through the Identity Store APIs.
Permission sets
A permission set is a template of policies plus a session duration. When you assign it to a user or group in an account, Identity Center creates an IAM role whose name starts with AWSReservedSSO_ in that account. Users assume that role through the portal or the CLI.
AWS managed, customer managed and inline policies
A permission set can combine three policy types:
- AWS managed policies such as
AdministratorAccess,PowerUserAccessorViewOnlyAccess. Nothing to deploy, and AWS updates them as services launch. - Customer managed policies, referenced by name. Each one must already exist in every target account under exactly that name and path.
- One inline policy of up to 32,768 bytes, of which at most 10,240 bytes can be non-whitespace.
The number of managed policies you can attach is bounded by the IAM quota on managed policies per role, which defaults to 10 and can be raised per account. Check the IAM Identity Center quotas page before you design around a higher number.
My default is AWS managed job-function policies plus an inline policy for exceptions. Customer managed policies pay off once infrastructure as code deploys them everywhere, because the name-matching rule turns manual drift into broken access.
For existing IAM roles with custom trust policies, role paths or ABAC role tags that a permission set can't reproduce, IAM's account access manager (an IAM feature, launched August 10, 2026) assigns those roles to Identity Center users and groups.
Session duration and permissions boundaries
Permission set session duration defaults to 1 hour, which is also the minimum, and goes up to 12 hours. Pick the shortest duration your engineers can work with, because a shorter session shrinks the window in which a stolen session is useful.
A permissions boundary caps what the role's policies can grant, which helps prevent privilege escalation through permission sets you let others edit. A customer managed boundary follows the same rule as customer managed policies: it must exist in each target account.
Multi-account access with AWS Organizations
With an organization instance, you assign groups to permission sets across any number of member accounts and change access in one place. Assign groups rather than individual users, so joiners and leavers only change group membership in your identity source.
Identity Center manages account access through its service-linked role, AWSServiceRoleForSSO, which it provisions in the member accounts it manages. Since August 5, 2026, new organization instances can leave AWS account access management switched off. Identity Center then provisions no service-linked role into member accounts, which suits teams that only want it for applications. You can switch it on later in the settings or with the UpdateInstance API.
Run Identity Center day to day from a delegated administrator account, not the management account. Our guide to delegated administrator best practices covers the setup and its limits. When an account joins or leaves the organization, check what happens to its IAM users and root access.
The AWS access portal and sessions
The AWS access portal is the page your users bookmark. After sign-in they see every account and permission set assigned to them, open the console for any of them, or copy short-term credentials for the CLI. Identity Center directory users also manage their own MFA devices there.
The classic portal URL ends in awsapps.com, and you can give it a custom alias. Since January 26, 2026, Identity Center also offers dual-stack endpoints for IPv6, with portal URLs ending in app.aws. With replication, each additional Region gets its own portal endpoints, while the awsapps.com URL and alias stay in the primary Region.
Two session settings matter. The permission set session duration (1 to 12 hours) controls how long role credentials last in an account. The user interactive session controls how long someone stays signed in to the portal: 8 hours by default, configurable from 15 minutes to 90 days. Don't stretch it toward the maximum just to save people a daily sign-in.
Security best practices: MFA and audit
Identity Center has few security settings, which makes them easy to get right and costly to skip.
Multi-factor authentication
For Identity Center directory users, MFA is on by default. The supported MFA types are FIDO2 authenticators (built-in ones such as Touch ID or Windows Hello, security keys such as YubiKey, and passkey providers), virtual authenticator apps that generate TOTP codes, and RADIUS MFA through AWS Managed Microsoft AD. Each user can register up to eight devices: two authenticator apps and six FIDO authenticators.
Prefer FIDO2 security keys or passkeys, because they're phishing-resistant and TOTP codes aren't. You can also require MFA when someone signs in from a new device or browser or an unknown IP address, and choose what happens when a user has no MFA device registered. For external IdP users, Identity Center's MFA doesn't apply, so enforce phishing-resistant MFA in the IdP.
Audit with CloudTrail
Identity Center actions are logged in CloudTrail in the Region where they happened. With multi-Region replication, sign-ins and changes can show up in more than one Region, so collect CloudTrail from every enabled Region, ideally through an organization trail. Because every person signs in as themselves, account activity traces back to a named user instead of a shared IAM user.
How to migrate off IAM users
Migrating is mostly sequencing. This is the order I'd run it in, from finding what exists to stopping it from coming back.
-
Inventory what exists. Generate the IAM credential report in each account and check when every access key was last used (
aws iam get-access-key-last-used). Turn on IAM Access Analyzer unused access findings across the organization. AWS Config rulesaccess-keys-rotatedandiam-user-unused-credentials-check, plus Security Hub controls IAM.3 and IAM.22, keep watching afterwards. -
Split people from workloads. People move to Identity Center. Workloads move to IAM roles: instance and function roles on AWS, IAM Roles Anywhere with X.509 certificates outside AWS, and
AssumeRoleWithWebIdentityfor OIDC sources such as GitHub Actions OIDC and Bitbucket Pipelines OIDC. CI/CD pipelines often hold the oldest keys, so start there. -
Rebuild access as permission sets. Group users by job function and attach the policies their IAM users or roles already used; the Identity Center FAQ confirms you can reuse them. Where a role depends on a custom trust policy, role path or ABAC role tags, assign it through IAM's account access manager instead.
-
Cut over by group. Assign the group, have members sign in through the portal and the CLI, and confirm they can work. Deactivate their access keys and passwords first, then delete the IAM users once last-used data shows nothing broke. Deactivation gives you a quick rollback; deletion doesn't.
-
Remove root credentials from member accounts. Enable centralized root access management in AWS Organizations, then delete root credentials from existing member accounts; new accounts start without them. The management account keeps its root user, which must have MFA. Our AWS account best practices cover the rest of that baseline.
-
Stop new IAM users appearing (our recommendation). An SCP that denies
iam:CreateUserandiam:CreateAccessKey, with exceptions for accounts that genuinely need IAM users, keeps the problem from returning. It's our pattern, not an AWS template: see the SCP that denies IAM user and access key creation. SCPs don't apply to the management account, so restrict who can sign in there. -
Keep one break-glass path (our recommendation). AWS advises break-glass access for privileged users in case the identity source is unavailable, but leaves the design to you. I recommend one documented emergency path for the organization, with phishing-resistant MFA, an alarm on every use, and a scheduled test.
Once people sign in through Identity Center, the next decision with long-term consequences is the Region.
Regions and multi-Region replication
You enable Identity Center in one primary Region, and users reach it from anywhere. Since February 3, 2026 you can replicate an organization instance to additional Regions, so a Regional outage no longer has to lock everyone out of their AWS accounts. The primary Region still holds the configuration you manage, which makes choosing it the bigger decision.
Choosing the primary Region
The primary Region holds all configuration, and replicas hold copies: users and groups, permission sets, applications and assignments. AWS's Region considerations list these factors:
- Where your users are, for latency
- Whether the Region is opt-in
- Whether the Region supports multi-Region replication
- Where you'll run AWS managed applications
- Digital sovereignty requirements
- Active Directory's home Region, which must match
- Cross-Region email: in some Regions, Identity Center sends user attributes such as email and name to Amazon SES in another Region
- AWS Control Tower, which creates the instance in your landing zone's Region
Opt-in Regions must be enabled in every account whose access you manage, new ones included, even if you never deploy there. AWS suggests an enabled-by-default Region to avoid that overhead, and I'd follow it unless sovereignty rules it out.
Changing the primary Region later means deleting the instance: disable AWS managed applications, recreate users, groups, permission sets, applications and assignments, and hand out a new portal URL. Snapshot assignments with the assignment APIs first. If the goal is a managed application in another Region, replicate instead.
Replicating to additional Regions
Replication launched for external IdPs in the 17 enabled-by-default commercial Regions. The Identity Center directory followed on July 29, and on October 1 support reached opt-in Regions and replication within AWS GovCloud (US) and within China. Active Directory sources aren't supported.
Before you can replicate to an additional Region, you need:
- An organization instance
- A multi-Region customer managed KMS key in the Identity Center account, with a replica key in each additional Region. KMS doesn't sync key policies, so copy the policy to each replica. Regions without a replica key don't appear in the console.
- Support for a customer managed key in every AWS managed application you use
Users, groups, memberships, permission sets, assignments, IdP SAML settings, trusted token issuers, application metadata and sessions replicate. Data inside AWS managed applications doesn't. You keep managing everything in the primary Region; additional Regions are mostly read-only apart from application management and session revocation. You can't promote an additional Region or remove the primary without deleting the instance, so pick an additional Region geographically far from the primary.
Excalidraw diagram loading.
Five details catch teams out:
- Each additional Region gets its own portal endpoints,
https://<instance-id>.<region>.portal.amazonaws.comand a dual-stackapp.awsequivalent. - If your IdP accepts multiple ACS URLs, add each Region's. IdP-initiated SSO still usually targets one, so point it at the primary and give users bookmark apps for the other Regions.
- If your IdP accepts only one ACS URL (AWS cites Google Workspace), additional Regions are backup only: during an outage you swap the ACS URL to another Region. Test that swap before you need it.
- Only tokens issued after you enable replication work across Regions, and CLI authorizations are per Region, so users authorize the CLI again in each one.
- Replication is asynchronous. An AWS walkthrough reported 15 to 30 minutes for the initial copy; treat that as an example, not a guarantee.
You can add a Region from the CLI and poll until its status is ACTIVE. Run both commands against the primary Region, adding --region <primary-region> if your CLI default differs:
aws sso-admin add-region --instance-arn <instance-arn> --region-name eu-central-1
aws sso-admin describe-region --instance-arn <instance-arn> --region-name eu-central-1
IAM Identity Center quotas and limits
Replication adds one quota to watch, AWS Regions enabled for one instance, which defaults to 6. The table covers the rest. These defaults come from the IAM Identity Center quotas page as of November 2026. Request increases from the management account or the delegated administrator account.
| Resource | Default | Adjustable |
|---|---|---|
| Permission sets per instance | 3,500 | Yes |
| Permission sets provisioned per AWS account | 500 | Yes |
| Inline policies per permission set | 1 | No |
| Inline policy size per permission set | 32,768 bytes (10,240 non-whitespace) | No |
| Users in the identity store | 1,000,000 | Yes |
| Groups in the identity store | 100,000 | Yes |
| AWS accounts configured | 7,000 | Yes |
| Applications configured (counted separately from accounts) | 7,000 | Yes |
| AWS Regions enabled for one instance | 6 | Yes |
| Groups per permission set per account, or per application | 100 | No |
| Connected Active Directory directories | 1 | No |
| Trusted token issuers | 10 | No |
| Instances per account | 1 | No |
A few limits sit outside the table. ProvisionPermissionSet with ALL_PROVISIONED_ACCOUNTS covers at most 3,500 accounts; beyond that, provision per account. Identity Center APIs share 20 TPS, applied per enabled Region. SCIM allows 25 write and 40 read TPS per API and runs only in the primary Region.
AWS recommends administering through the CLI or APIs once you pass 50,000 users, 10,000 groups, 500 permission sets or 3,000 applications. Long before that, I'd manage permission sets and assignments as infrastructure as code.
What changed in 2026
If you set up Identity Center before this year, these are the changes worth checking against your design:
- January 26: IPv6 through dual-stack endpoints (portal
app.aws, adminapi.aws). Taipei and AWS GovCloud (US) followed on February 27. - February 3: multi-Region replication for external IdPs. The Identity Center directory followed on July 29, a one-click multi-Region option for new organization instances on August 7, and opt-in Regions, GovCloud (US) and China on October 1.
- June 30: programmatic AWS account access for customer managed applications that use an external IdP as a trusted token issuer.
- August 5: managing AWS account access became optional for new organization instances.
- August 10: account access manager, an IAM feature, assigns existing IAM roles to Identity Center users and groups.
- October 5: network access controls for the Identity Store and SCIM APIs: source IP ranges for both, and allowed VPC endpoints for the Identity Store API.
- October 19: optimistic locking for identity store users and groups.
- October 20: changing the identity source away from Active Directory keeps users, groups, memberships and assignments.
- October 26: cross-Region trusted identity propagation.
If you're starting fresh, here's the order to set it up.
Getting started with IAM Identity Center
Settle four things before you enable anything: AWS Organizations with all features enabled, access to the management account, your identity source, and your primary Region, including whether you need multi-Region and therefore a customer managed KMS key.
- In the management account, open the IAM Identity Center console in your chosen Region and enable an organization instance. In Regions that support multi-Region, you choose single-Region, multi-Region or custom. The multi-Region option creates the KMS key and replicates to one additional Region; your principal needs
kms:CreateKey,kms:TagResourceandkms:ReplicateKey. - Keep AWS account access management on if people need access to accounts.
- Register a delegated administrator account and do the rest from there. For Active Directory, AWS recommends placing the directory in that account; a directory in the management account has to be connected from the management account, because the delegated administrator lacks the permissions.
- Connect your identity source: create users in the directory, connect Active Directory, or set up SAML and SCIM with your IdP.
- Create a small set of permission sets (admin, developer, read-only), assign them to groups per account, and pilot with one team.
Developers then set up the AWS CLI with IAM Identity Center using aws configure sso. SDKs reuse the same profile, as shown when you configure Boto3 with IAM Identity Center.
ABAC and trusted identity propagation
Attribute-based access control (ABAC) lets one permission set serve many teams. You pass user attributes such as department or cost center from your identity source, write policies that compare aws:PrincipalTag with resource tags, and tag resources to match. You end up with fewer permission sets, and access follows attribute changes in your identity source.
Trusted identity propagation passes the user's identity, not only a role, to AWS analytics services such as Amazon Redshift, Amazon S3 Access Grants, Amazon EMR and AWS Lake Formation. Data access can then be granted and audited per user. Since October 26, 2026 it also works across Regions.
Troubleshooting common problems
When access breaks, start with the check that matches the symptom.
Users or groups don't sync from the IdP
Check the SCIM token first; an expired token stops synchronization entirely. Then confirm the user has all four required name attributes, no multivalue attributes, and an assignment to the AWS application in the IdP. The IdP controls the sync interval.
A user can't reach an account
Confirm an AWSReservedSSO_* role exists in the target account, the permission set is assigned to that account and the user's group, and the user is in that group. Check that any customer managed policies exist in that account.
MFA enrollment or CLI sign-in fails
A user who already has eight MFA devices can't add another. External IdP users enroll MFA in the IdP. For CLI errors, see the fixes for ForbiddenException when calling GetRoleCredentials and missing the following required SSO configuration values.
Frequently Asked Questions
Is AWS IAM Identity Center free?
Is IAM Identity Center the same as AWS SSO?
Do I need AWS Organizations to use IAM Identity Center?
Does IAM Identity Center replace IAM?
Can I change the IAM Identity Center Region later?
Can I switch from Active Directory to Okta or Entra ID without losing assignments?
Wrapping up
Run AWS IAM Identity Center as an organization instance, for people, and use IAM roles for workloads. Choose the primary Region deliberately, and if outage resilience matters, plan multi-Region and its KMS key from day one. Rotate the SCIM token well before its one-year expiry.
Start with the IAM credential report and Access Analyzer's unused access findings, then pilot with one group. After that, set up the AWS CLI for your engineers and move administration to a delegated administrator account.
Next step