Skip to content
Cloud Security Risk Assessment

Cloud security maturity model: five levels for the cloud access layer

The cloud security maturity model here has five levels for the access layer: Unmanaged, Reactive, Defined, Managed and Optimized. Each level is a band of the 0 to 100 domain score from the 40-question assessment and lists criteria a reviewer can observe, so two people reach the same level from the same evidence.

Start the assessment ยท All guides

Five levels map to bands of the domain score

The model covers the access layer only. A team scores each of the six domains in the assessment and reads the level from the band. The bands are this site's reading convention for the 0 to 100 scale, not a published standard.

LevelNameDomain score bandObservable criteria
1Unmanaged0 to 20Local human users with passwords exist; admin ports are open to the internet; audit logs are on in some accounts only; nobody owns the access list.
2Reactive21 to 40Root MFA and SSO are in place; public storage is blocked in the main accounts; fixes follow incidents; no review calendar exists.
3Defined41 to 60Written standards cover roles, keys and logging; reviews happen every 90 days; organization-level deny policies exist for a few actions.
4Managed61 to 80Privileged roles are granted for limited time; drift is reported within a day; exceptions expire; alerts reach a person.
5Optimized81 to 100Revocation and evidence drills run on schedule; guardrails are tested before enforcement; AI agents have their own identity and approvals.

A maturity level is read per domain, not once for the account

A team can reach level 4 in evidence and stay at level 2 in identities. The lowest weighted domain usually decides the real exposure, so report all six levels. The domain weights are 25, 20, 20, 15, 10 and 10, and identities and entitlements carries the largest.

An overall level is optional. If you need one, compute the weighted average of the six domain scores and apply the same bands.

Criteria are things a reviewer can see

Each criterion names an artifact: a credential report, a policy list, a calendar entry, an alert rule. Words such as "mature" or "good" are not criteria. Level 3 in identities, for example, needs a written role standard and a review every 90 days with removals recorded.

Level 4 adds time-limited grants. The check is no standing privilege, which weighs 3 and is rated high effort, so many teams stall between levels 3 and 4.

How the model relates to other published models

The CSA publishes the Cloud Controls Matrix, a control framework with domains and control specifications (CSA, read 16 Nov 2026). It is a list of controls, not a five-level scale. The model on this page borrows no text from it and uses its own bands.

The CIS Benchmarks give configuration checks per cloud and are named on each question page. Use them as the test for a level, and use this page to turn the results into a level.

A reviewer asks for the same artifacts at every level

The reviewer asks for the artifact behind each criterion before awarding the level. For level 3 in guardrails, that is the list of organization-level deny policies. For level 4, it is a drift report with dates inside the last day, and the exception register with expiry dates.

If the artifact is missing, the criterion fails and the level drops by one. This keeps the model from becoming a matter of opinion.

The levels read differently in each domain

In AI workloads, level 1 means agents run under a shared or personal credential. Level 3 means each agent has its own identity and a written action list, as in own identity and tool permissions. Level 5 adds human approval for high-impact actions and logging of every call.

In evidence, level 1 is audit logs on in some accounts. Level 5 is a revocation drill that cut off one identity within an hour and a record of the result.

Moving up one level takes the highest weighted gaps first

To go from level 2 to level 3 in a domain, fix the low-effort checks that weigh 3: root account MFA, open admin ports, public storage and audit logging. The methodology explains how the five fixes are chosen, and the options page lists tools that report on these checks.

Sources

  1. CSA Cloud Controls Matrix, read 16 Nov 2026
  2. CIS Benchmarks, read 16 Nov 2026
  3. NIST SP 800-53 Rev. 5, read 16 Nov 2026

Last reviewed 16 Nov 2026.