# Cloud security assessment checklist: 40 checks for the cloud access layer

Facts read 16 Nov 2026. Page reviewed 16 Nov 2026.

The cloud security assessment checklist lists 40 checks in six domains: identities and entitlements (8), network reachability (7), data perimeter (7), AI workloads (6), guardrails and drift (6) and evidence (6). Each applies to AWS, Azure and Google Cloud and weighs 1 to 3 in the score. Edition 16 Nov 2026.

## Forty checks sit in six domains, and the domains carry unequal weight

The checklist has 40 checks. Each is a yes, partly or no question about the cloud access layer, and each applies to AWS, Azure and Google Cloud. A check weighs 1, 2 or 3 depending on how far a wrong answer moves risk. The 40 weights add up to 84, and 12 checks weigh 3.

Domain weights are set apart from the check weights: identities and entitlements 25, network reachability 20, data perimeter 20, AI workloads 15, guardrails and drift 10, evidence 10. The domain weights add up to 100.

| Domain | Checks | Domain weight | Sum of check weights |
| --- | --- | --- | --- |
| Identities and entitlements | 8 | 25 | 19 |
| Network reachability | 7 | 20 | 14 |
| Data perimeter | 7 | 20 | 15 |
| AI workloads | 6 | 15 | 14 |
| Guardrails and drift | 6 | 10 | 11 |
| Evidence | 6 | 10 | 11 |

## Eight identity checks cover who holds power and for how long

People and workloads that hold standing power over the cloud account are the most common route from one stolen credential to a full breach.

- [Root or global administrator has phishing-resistant MFA and no keys](/questions/root-account-mfa/) (weight 3, fix effort low)
- [People sign in through single sign-on with MFA](/questions/workforce-sso-mfa/) (weight 3, fix effort medium)
- [Long-lived keys removed or rotated every 90 days](/questions/long-lived-keys/) (weight 3, fix effort medium)
- [Workloads use platform-issued identities](/questions/workload-identity/) (weight 2, fix effort high)
- [Administrator roles are few, named and separate](/questions/admin-roles-limited/) (weight 2, fix effort low)
- [Privileged roles are granted for a limited time](/questions/no-standing-privilege/) (weight 3, fix effort high)
- [Permissions reviewed every 90 days](/questions/access-reviews/) (weight 2, fix effort medium)
- [Emergency accounts tested twice a year](/questions/emergency-access/) (weight 1, fix effort low)

## Seven network checks decide what an attacker can touch without a credential

Reachability decides which of your services an attacker can touch before any credential is needed.

- [No admin or database port open to the internet](/questions/no-open-admin-ports/) (weight 3, fix effort low)
- [Data services reached through private endpoints](/questions/private-service-endpoints/) (weight 2, fix effort high)
- [No public IP address by default](/questions/no-default-public-ip/) (weight 2, fix effort medium)
- [Admin access brokered and logged](/questions/brokered-admin-access/) (weight 2, fix effort medium)
- [Metadata service hardened](/questions/metadata-service-hardened/) (weight 3, fix effort low)
- [Outbound traffic filtered to known destinations](/questions/egress-controlled/) (weight 1, fix effort high)
- [Reachability analyzed monthly](/questions/reachability-analysis/) (weight 1, fix effort medium)

## Seven data perimeter checks limit who reads data and from where

A data perimeter limits who can reach your data and from where, even when an identity or a policy is wrong.

- [Public object storage blocked organization-wide](/questions/public-storage-blocked/) (weight 3, fix effort low)
- [Sharing limited to your own organization](/questions/organization-boundary/) (weight 3, fix effort medium)
- [Sensitive data readable from trusted networks only](/questions/trusted-network-condition/) (weight 2, fix effort high)
- [External access findings reviewed weekly](/questions/external-access-findings/) (weight 2, fix effort low)
- [Customer-managed keys with limited use](/questions/key-control/) (weight 2, fix effort medium)
- [Sensitive data locations known](/questions/sensitive-data-located/) (weight 1, fix effort medium)
- [Third-party roles conditioned, owned and dated](/questions/third-party-access/) (weight 2, fix effort medium)

## Six AI checks treat each agent as an identity with a reach

AI services and agents act with cloud credentials, so what they can call and read is an access question.

- [Each AI agent has its own identity](/questions/ai-own-identity/) (weight 3, fix effort medium)
- [Agent actions limited to the task and written down](/questions/ai-tool-permissions/) (weight 3, fix effort high)
- [Only approved models can be used](/questions/approved-models-only/) (weight 2, fix effort low)
- [AI data reach matches user rights](/questions/ai-data-reach/) (weight 2, fix effort high)
- [Human approval for high-impact agent actions](/questions/human-approval/) (weight 2, fix effort medium)
- [Model and tool calls logged with the identity](/questions/ai-activity-logging/) (weight 2, fix effort low)

## Six guardrail checks stop bad changes and catch the ones that slip past

Preventive controls stop a bad change before it lands, and drift checks catch the ones that slip past.

- [Deny policies at the organization level](/questions/preventive-policies/) (weight 3, fix effort medium)
- [Changes go through code and pipelines](/questions/changes-as-code/) (weight 2, fix effort high)
- [Drift reported within a day](/questions/drift-detection/) (weight 2, fix effort medium)
- [Guardrails tested before enforcement](/questions/guardrail-testing/) (weight 1, fix effort medium)
- [Exceptions have owner, reason and expiry](/questions/exceptions-expire/) (weight 1, fix effort low)
- [Guardrails protected from tampering](/questions/tamper-protection/) (weight 2, fix effort medium)

## Six evidence checks let you prove access and cut it off

Logs and records are what let you prove who had access, detect misuse and respond.

- [Audit logs on in every account and region](/questions/audit-logs-everywhere/) (weight 3, fix effort low)
- [Logs stored where workload admins cannot change them](/questions/logs-protected/) (weight 2, fix effort medium)
- [Logs kept 365 days, 90 searchable](/questions/log-retention/) (weight 2, fix effort low)
- [Identity alerts reach a person](/questions/identity-alerts/) (weight 2, fix effort medium)
- [Access evidence produced in one working day](/questions/access-evidence-on-demand/) (weight 1, fix effort medium)
- [One identity cut off within one hour](/questions/revocation-drill/) (weight 1, fix effort high)

## The checklist is also a Word file you can edit

The Word version, [cloud-security-assessment-checklist](/cloud-security-assessment-checklist/), lists the same 40 checks with columns for the answer, the owner and the evidence. Use it when a team wants to review the checks in a meeting before anyone opens the scored tool.

To get a score, answer the same checks in the [assessment](/). The scoring is explained in the [methodology](/methodology/), and the [risk register template](/risk-register-template/) turns the five fixes into tracked work. Providers publish more checks than these 40, as the [options](/options/) page shows.

## Each check names the control it comes from

Every check cites one primary control: a CIS Benchmark, the Microsoft cloud security benchmark, a provider guide or NIST SP 800-53 Rev. 5. CIS says it publishes more than 100 Benchmarks across 25 or more vendor product families (CIS Benchmarks page, read 16 Nov 2026). The checklist refers to CIS and CSA controls by number and short title only, because those texts are licensed.

## Download

- [Cloud security assessment checklist (.docx)](/downloads/cloud-security-assessment-checklist.docx): Word document made from this page, edition 16 Nov 2026

## Related pages

- [AWS Security Checklist for the Access Layer](/aws-security-checklist/)
- [Azure Security Checklist for the Access Layer](/azure-security-checklist/)
- [Google Cloud Security Checklist: Access Layer](/google-cloud-security-checklist/)
- [All of them](/guides/)

## Sources

- CIS Benchmarks: https://www.cisecurity.org/cis-benchmarks (read 16 Nov 2026)
- NIST SP 800-53 Rev. 5: https://csrc.nist.gov/pubs/sp/800/53/r5/upd1/final (read 16 Nov 2026)
- Microsoft cloud security benchmark overview: https://learn.microsoft.com/en-us/security/benchmark/azure/overview (read 16 Nov 2026)
- AWS Well-Architected Security Pillar: https://docs.aws.amazon.com/wellarchitected/latest/security-pillar/welcome.html (read 16 Nov 2026)

Last reviewed 16 Nov 2026. Canonical: https://cloudsecurityriskassessment.com/cloud-security-assessment-checklist/
