A Six-Section IAMOps Audit Checklist You Can Run in an Afternoon

A Six-Section IAMOps Audit Checklist You Can Run in an Afternoon

Owners of light sport aircraft know the value of a written pre-flight list: it turns “I think I checked” into a record. The same discipline applies to the policies that govern who and what can act inside your software systems. This checklist converts the ideas the Ops Atlas groups under IAMOps, an informal label for managing authorisation as code, into audit steps you can work through in one sitting and repeat on a schedule. Mark each item pass, fail or not applicable, and keep the sheet.

Running the audit step by step

Set aside two hours and a quiet room. Gather read access to your policy repository or console, a list of service accounts and a month of access logs. The wording of each point is ours, but the underlying criteria come from the Ops Atlas page on IAMOps, which also explains why each matters. Where a control family is mentioned, it refers to NIST SP 800-53 Revision 5. Work top to bottom and resist the urge to fix items mid-audit; a clean picture of the current state is worth more than a half-repaired one. Two people make a better audit than one, because the author of a policy tends to read what they meant instead of what is written. If an item needs data you cannot reach, mark it fail for now and note who can supply it.

Section A: Ownership and inventory

  • Every role, service account and third-party integration has a named human owner.
  • Each owner can state in one sentence why the identity exists.
  • A list exists of identities that can be assumed from outside your organisation, and someone has reviewed it.
  • Orphaned identities (owner gone, purpose unknown) are tagged for removal, not left alone.

An inventory is the first rung because nothing else can be measured without it. As with a maintenance log, an unrecorded component cannot be inspected.

Section B: Grants and ceilings

  • For your five most sensitive actions, you can name the policy that grants each and the policies that cap it.
  • No organisation-level or boundary policy is relied upon to grant access.
  • Explicit denies are listed in one place, with a reason beside each.
  • No statement combines Allow with NotAction without a written justification.

These items reflect the Atlas’s point that some policy types only limit. They connect to the least-privilege family, AC-6, in the NIST catalogue.

Section C: Measuring excess

  • Permission utilisation is calculated per role (actions used divided by actions allowed). The Atlas example scored one role at 15 percent, with 27 of 180 used.
  • The share of allow statements containing wildcards is known. In the Atlas example it was 168 of 1,400, about 12 percent.
  • Roles scoring below your threshold have a documented follow-up date.
  • The audit records known blind spots in activity data, namely seasonal jobs, permissions that pass a role to another service, and calls made straight to stored data.

The review of user privileges described in control AC-6(7) is served by this section, since it asks that privileges be reviewed periodically and reassigned or removed when needed.

Section D: Credentials

  • The percentage of credentials older than ninety days is measured and trending downward.
  • At least the most critical deployment pipeline authenticates through federation rather than a stored key.
  • Trust conditions name exact, per-environment subjects and contain no wildcards that admit a whole organisation of repositories.
  • Any legacy test application with elevated access has been located and either removed or restricted.

The last item is worth underlining: forgotten test apps with broad rights have featured in widely reported incidents, so seek them out deliberately.

Section E: Testing the rules themselves

  • Policy changes pass through version control and automated checks; the Atlas’s third maturity level expects at least 90 percent of changes to follow this route.
  • Tests include requests with missing context keys, because negated conditions can match an absent value.
  • Policies are validated against a schema where the engine supports one, since some engines skip a policy that errors and so leave a protective rule inactive.
  • Exemptions based on a role name are paired with a deny that protects the name itself.

Similar thoroughness appears in our walkthrough of gear inspection points for LSA owners, where each step has an expected result written beside it. Give each audit item the same treatment: an expected state, not just a tick.

Section F: Logging and review

  • Use of privileged functions is logged and the log is read by someone (compare control AC-6(9)).
  • A monthly review is on the calendar, with attendees and a short agenda.
  • An exception register lists every deliberately broad permission with an owner and an expiry date.
  • Access refusals after each release are counted, and time-to-grant is tracked so friction is visible.

The Atlas treats a rise in grant time, from 36 to 48 hours in its example, as acceptable when exposure falls. What matters is that the trade is noticed and agreed upon.

Scoring your sheet

Count the passes. Fewer than a third suggests you are at the Atlas’s first level, ad hoc. Roughly half indicates a guarded posture with no new long-lived keys. Most items passing, with changes flowing through automation and a regular review, places you around the codified and measured levels. Treat the result as a conversation starter, not a certificate.

Fuel handling gives a useful parallel. Our note on common fuel system errors for sport pilots shows how small oversights compound, which is exactly why a recurring audit beats a one-time cleanup. Fix the failed items by risk, date the sheet, and schedule the next pass before you close the file.