IAM least privilege works the same in three clouds and is configured differently in each
Facts read 16 Nov 2026. Page reviewed 16 Nov 2026.
IAM least privilege is the practice of limiting what identity and access management lets each identity do. AWS, Azure and Google Cloud each document a narrow role, a scope, a usage report and a ceiling, under different names. This page lines the four up, with documentation read on 16 Nov 2026.
Open the calculator ยท The field guide
The same four controls exist in each cloud
Each provider offers a way to pick a narrow role, to limit where it applies, to see what was used, and to cap what grants can allow. The names differ, and so do the traps, so a control that works in one cloud cannot be assumed to behave the same in another.
| Control | AWS | Azure | Google Cloud |
|---|---|---|---|
| Narrow role | Customer managed policy or job function policy | Built-in or custom Azure role | Predefined or custom role |
| Scope | Resource ARNs and conditions in the policy | Management group, subscription, resource group, resource | Organization, folder, project, resource |
| Usage report | Last accessed information, Access Analyzer unused access | Defender for Cloud permissions management | Role recommendations |
| Ceiling | Permissions boundary, service control policy | Deny assignment | Deny policy, Principal Access Boundary |
AWS narrows by policy and reports by last use
AWS puts the limits in the policy: actions on specific resources under specific conditions. Its job function managed policies are a starting point, and Access Analyzer can generate a policy from CloudTrail activity. The last accessed information shows service and management action use with a tracking period of at least 400 days for service information, and Access Analyzer unused access findings list unused roles, access keys, passwords and permissions.
AWS says to regularly review and remove unused users, roles, permissions, policies and credentials. For the ceiling, permissions boundaries limit one user or role, and service control policies limit the accounts in an organization. Because the console shows one account at a time, run the reports at the organization level where the provider allows it, so that a forgotten account is not the one that holds the oldest role. The identity types are covered in AWS human identities and AWS service identities.
Azure narrows by role and scope and reports through Defender for Cloud
Azure RBAC grants a role at a scope. Microsoft's guidance is to limit subscription owners to a maximum of 3, to avoid broader roles at broader scopes, and to use a narrow scope when a privileged administrator role must be assigned. Privileged Identity Management adds time-bound access, and access reviews in Entra ID repeat the check on a schedule.
Microsoft says Defender for Cloud analyzes identity configurations and usage patterns to identify excessive, unused or misconfigured permissions for users, service principals, managed identities and service accounts. The ceiling is the deny assignment, which blocks actions even when a role assignment grants them. See Azure human identities and Azure workload identities.
Google Cloud narrows by role and reports through recommendations
Google separates basic roles, which hold thousands of permissions, from predefined roles and custom roles. Role recommendations compare granted permissions with those used in the last 90 days and suggest removing or replacing a role, and Policy Simulator shows the effect of a change before it is made.
Google recommends dedicated service accounts for each application and warns about default service accounts that carry the Editor role. Its ceilings are deny policies, which apply regardless of granted roles, and Principal Access Boundary policies, which define the resources a set of principals can access. See Google Cloud service identities and Google Cloud third-party identities.
Three limits are shared by every provider report
Read the reports with the same care in every cloud. All three reports are usage-based, so a role used less often than the window looks unused. None of the three applies a removal on its own: you apply the change, and you answer for its effect. And none sees access that is granted through a path outside IAM, such as a key stored in a secret manager.
Treat the reports as a list of candidates, then test each change. Record the date each report was run, the window it covered and the person who accepted or dismissed each item, because the same report run next quarter will show different results as workloads change.
The mechanisms behind the ceilings are on least privilege access control, and ten before and after settings are in least privilege examples.
Related pages
Sources
- AWS, security best practices in IAM, read 16 Nov 2026
- AWS, managed policies for job functions, read 16 Nov 2026
- AWS, refine permissions using last accessed information, read 16 Nov 2026
- AWS, IAM Access Analyzer findings, read 16 Nov 2026
- AWS, permissions boundaries for IAM entities, read 16 Nov 2026
- AWS, service control policies, read 16 Nov 2026
- Microsoft, best practices for Azure RBAC, read 16 Nov 2026
- Microsoft, what is Privileged Identity Management, read 16 Nov 2026
- Microsoft, permissions management in Defender for Cloud, read 16 Nov 2026
- Microsoft, list Azure deny assignments, read 16 Nov 2026
- Google Cloud, using IAM securely, read 16 Nov 2026
- Google Cloud, role recommendations overview, read 16 Nov 2026
- Google Cloud, best practices for service accounts, read 16 Nov 2026
- Google Cloud, deny policies, read 16 Nov 2026
- Google Cloud, Principal Access Boundary policies, read 16 Nov 2026
Last reviewed 16 Nov 2026.