A cloud IAM access review gives your team a repeatable way to find excessive permissions, confirm ownership, and remove unused access without putting production systems at unnecessary risk. This runbook covers human users, roles, service accounts, API keys, privileged access, review cadence, and a practical tracker you can reuse for quarterly audits or after organizational change.
Overview
Cloud identity and access management (IAM) tends to grow alongside an application. A developer joins a project, a contractor receives temporary access, a deployment pipeline creates a service account, and an incident leads to a broad emergency permission. Unless those decisions are revisited, access can remain long after the original need has disappeared.
A cloud permissions audit is not simply a list of users. It should connect four questions for every identity and permission set:
- Who or what has access?
- What can it do?
- Why is that access still required?
- Who owns the decision, and when will it be reviewed again?
The exact names differ between providers. One platform may use roles and policies, while another separates groups, bindings, service principals, or workload identities. The review method remains broadly applicable: inventory identities, map effective permissions, validate business need, reduce unnecessary privilege, and retain evidence of the decision.
Use this runbook alongside your broader cloud security checklist for developers. IAM reviews are one control within a larger security and operations program.
What to track
1. Human users and groups
Start with the identity provider and each cloud account, subscription, project, or organization boundary. Record active employees, contractors, administrators, break-glass accounts, and users who authenticate through a federated identity provider. Group membership matters as much as direct assignment because inherited access can be easy to miss.
For each person, capture:
- Identity and employment or contract status
- Team, manager, and application or environment owned
- Groups, roles, and direct permissions
- Last successful sign-in, where available
- Whether production access is required
- Access owner and approval date
- Expiration or next-review date
Do not treat a recent sign-in as proof that every permission is necessary. A user may need occasional production access but not unrestricted administration.
2. Privileged roles
Review organization-wide administrators, security administrators, billing administrators, identity administrators, and any custom role that can change access controls or production infrastructure. Confirm that each assignment has a named owner and a documented operational reason.
Where the platform supports it, prefer time-limited elevation or just-in-time access for high-impact actions. Separate routine development permissions from permissions that can alter identity policy, delete data, modify network boundaries, or disable logging. Your objective is not to eliminate all privileged access; it is to make persistent privilege deliberate and visible.
3. Service accounts and workload identities
Service account security requires understanding how applications use identities. Inventory accounts used by deployment pipelines, scheduled jobs, containers, serverless functions, database migration tools, monitoring agents, and AI or data-processing workloads.
For each workload identity, document:
- Application, repository, job, or cluster that uses it
- Environment and cloud resources it can reach
- Human or team owner
- Authentication method and credential location
- Permissions granted directly or through attached roles
- Last observed use and expected rotation or review date
Prefer workload identity mechanisms that avoid long-lived keys when your provider and application design support them. A service account with no clear owner is a remediation priority, even if its permissions appear narrow.
4. API keys, secrets, and dormant credentials
Search for API keys, access keys, tokens, certificates, and secrets associated with both human and machine identities. Record creation dates, last-use information, storage location, allowed services, and rotation owner. A key that has not been used recently may be removable, but verify logs and application dependencies first.
Never paste secret values into the access-review spreadsheet. Store only an identifier or fingerprint that lets an authorized operator locate the credential in the approved secrets system.
5. A reusable review tracker
A spreadsheet or ticket queue can provide sufficient control for a small team if it is maintained consistently. Use one row per identity-to-role relationship, or one row per credential when key-level review is required.
| Field | Example entry |
|---|---|
| Identity | developer@example.com or payments-deploy |
| Type | User, group, service account, key |
| Environment | Development, staging, production |
| Permission or role | Named provider role or custom policy |
| Business purpose | Deploy API service to production |
| Owner | Team or accountable individual |
| Last-use evidence | Sign-in, audit event, pipeline execution, or unknown |
| Decision | Keep, reduce, suspend, remove, investigate |
| Approval and date | Reviewer, ticket, and completion date |
| Next review | Scheduled date or event trigger |
Cadence and checkpoints
Set the review schedule according to risk and change rate rather than choosing one universal interval. A quarterly review is a practical baseline for many teams, while sensitive production environments or frequently changing organizations may need more frequent checks. The schedule should be written down and assigned to an owner.
Monthly operational checks
- Review newly created users, roles, service accounts, and credentials.
- Check failed or unusual authentication activity through available logs.
- Confirm that offboarding and contractor end dates were processed.
- Review newly granted privileged access and temporary exceptions.
Quarterly access review
- Export or query the current identity and permission inventory.
- Ask each application or team owner to confirm access.
- Compare assigned permissions with observed usage where data is available.
- Remove dormant users, expired exceptions, and unowned credentials after validation.
- Record approvals, decisions, and unresolved risks.
Event-driven reviews
Run an additional review after an employee or contractor leaves, a team changes ownership, an acquisition or migration occurs, a major incident happens, or a new production environment is introduced. Include IAM changes in deployment and infrastructure change procedures, including Terraform or other infrastructure-as-code workflows.
Before removing access, identify dependencies. A service account may be used by a scheduled job that runs infrequently, and a human user may be the only operator who can complete a recovery procedure. Coordinate changes, use a ticket or change record, and define a rollback path for production permissions.
How to interpret changes
Access-review results are useful when they lead to decisions, not just a larger inventory. Classify each finding into a clear action:
- Keep: The permission is required, owned, and supported by current evidence.
- Reduce: The identity needs the capability, but a narrower role, resource scope, or environment boundary is appropriate.
- Time-limit: The access is valid for a project, incident, or migration and should have an explicit expiration.
- Suspend and verify: The access appears dormant or unowned, but application dependency checks are still needed.
- Remove: The identity, role assignment, or credential has no current business purpose.
- Investigate: The team cannot explain the access, its use differs from expectations, or audit evidence is incomplete.
Pay particular attention to mismatches: a development identity with production administration, a deployment account with broad access across unrelated applications, or a user who has accumulated multiple overlapping roles. These patterns do not automatically indicate misuse, but they are good prompts for redesign.
Track the trend over time. Useful measures include the number of unowned identities, persistent privileged assignments, credentials past their review date, expired exceptions, and unresolved findings. Avoid optimizing for a zero count at the expense of operational safety. A documented exception with an owner and expiration is more manageable than undocumented access that appears clean only because it was deleted without dependency analysis.
If an access change could affect backups, databases, or recovery operations, coordinate it with your disaster recovery testing and restore runbook. IAM failures can become recovery failures when operators cannot reach required systems.
When to revisit
Make the access review a recurring operational task rather than a one-time compliance exercise. Schedule the next review before closing the current one, and preserve the tracker, evidence, and decisions in a location the next reviewer can use.
Revisit this runbook at least quarterly, and update it sooner when your provider changes IAM terminology or console workflows, your identity provider integration changes, new workload-identity features are adopted, or your organization introduces a new environment. Review the procedure after an access-related incident and after major architecture changes such as a Kubernetes migration, a multi-tenant redesign, or the addition of AI infrastructure. Related architecture decisions may affect which teams need access to clusters, databases, object storage, vector databases, or model-serving systems; document those boundaries explicitly.
For the next review, take these actions:
- Assign one accountable reviewer and one owner for every production application.
- Export the current identity, role, and credential inventory.
- Populate the tracker with purpose, owner, evidence, decision, and next-review date.
- Review privileged access first, then service accounts and API keys, then routine user access.
- Validate dependencies before disabling or removing anything.
- Apply approved reductions and record the change references.
- Schedule the next monthly checks, quarterly review, and relevant event-driven triggers.
A consistent cloud IAM access review makes permissions easier to explain, safer to change, and simpler to maintain as your application and team grow.