Almost every cloud audit, cyber insurance questionnaire and CSPM dashboard asks whether you meet "the CIS benchmark", yet few teams can say which benchmark, which version or which profile they are measured against. The Center for Internet Security (CIS) publishes foundations benchmarks for eight cloud providers, separate service benchmarks for compute, database and storage, and managed Kubernetes benchmarks, and the scanners built into each cloud often assess an older version than the one CIS currently publishes. This guide explains what the cloud foundations benchmarks cover, how they are structured, where identity and machine credentials feature, how to adopt and scan against them, and how to produce evidence an assessor will accept. It is for cloud security engineers, compliance leads and auditors. For the CIS Controls themselves see the CIS Controls Guide, and for benchmarks in general the CIS Benchmarks Guide. Versions below were checked on cisecurity.org on 27 October 2026.
Key takeaways
- Name the version. "CIS compliant" means nothing without the benchmark, version and profile, for example CIS Amazon Web Services Foundations Benchmark v7.1.0 Level 1.
- Native scanners lag. On the date checked, AWS Security Hub CSPM recommended CIS AWS v5.0.0 while CIS published v7.1.0; Microsoft Defender for Cloud listed Azure v3.0 against a current v6.0.0; Google Compliance Manager offered GCP v3.0 against v5.0.0.
- Foundations are not everything. The foundations benchmarks cover account-level settings; compute, database, storage and Kubernetes have their own benchmarks.
- Identity carries most of the weight. Root and admin protection, MFA, unused credentials, access key rotation and service account keys appear in every cloud's benchmark.
- Manual recommendations still count. Automated scans cannot evaluate some recommendations, so assessors expect documented evidence for those.
What it is and who it applies to
CIS Benchmarks are consensus-based secure configuration guides. Volunteers, technology vendors, community members and CIS staff draft them on CIS WorkBench, discuss them until the community reaches consensus, and CIS publishes the result. CIS states the benchmarks are freely available in PDF format for non-commercial use; CIS SecureSuite members also get machine-readable formats such as XCCDF, Build Kits and CIS-CAT Pro, and tailoring requires membership. CIS also publishes hardened virtual machine images on cloud marketplaces.
No law makes a CIS benchmark mandatory by itself, but they apply in practice when a contract, an auditor, a regulator's guidance or your own policy names them, and when your CSPM reports against them. NIST SP 800-53 CM-6 (Configuration Settings) expects organisations to establish configuration settings using common secure configurations, and CIS Benchmarks are a common way to meet that.
The cloud foundations family
| Provider | Foundations benchmark (current version) | Related CIS benchmarks | Native assessment, as documented |
|---|---|---|---|
| Amazon Web Services | AWS Foundations v7.1.0 | AWS Compute Services 2.0.0, Database Services 2.0.0, End User Compute Services 1.2.0, Storage Services 1.0.0; EKS and EKS Auto Mode | Security Hub CSPM: v5.0.0 (recommended), v3.0.0, v1.4.0, v1.2.0 |
| Microsoft Azure | Azure Foundations v6.0.0 | Azure Compute Services 2.0.0, Database Services 2.0.0, Storage Services 2.0.0; AKS and AKS Automatic | Defender for Cloud: CIS Azure Foundations v2.1.0 and v3.0 |
| Google Cloud | Google Cloud Platform Foundation v5.0.0 | Container-Optimized OS 1.2.0; GKE and GKE Autopilot | Security Command Center Compliance Manager: CIS Google Cloud Computing Platform Benchmark v3.0 (Premium) |
| Oracle Cloud Infrastructure | OCI Foundations v3.1.1 | Oracle SaaS Cloud Applications 1.0.0; OKE | Not reviewed for this guide |
| Alibaba Cloud | Alibaba Cloud Foundation v2.0.1 | None listed | Not reviewed |
| IBM Cloud | IBM Cloud Foundations v2.0.0 | None listed | Not reviewed |
| DigitalOcean | DigitalOcean Foundations v1.0.1 | DigitalOcean Services 1.0.0 | Not reviewed |
| Tencent Cloud | Tencent Cloud Foundation v1.0.1 and Tencent Cloud Enterprise Foundation v1.0.0 | TDSQL benchmark | Not reviewed |
Microsoft 365 and Google Workspace have their own foundations benchmarks; they are SaaS rather than cloud infrastructure and are out of scope here. See the Microsoft 365 Security Guide.
Structure and key requirements
- Profiles: CIS defines Level 1 as a base recommendation that can be implemented quickly with little performance impact, and Level 2 as defence in depth for environments where security is paramount, which can adversely affect operations if applied carelessly. Some benchmarks also have a STIG profile, which replaced the former Level 3.
- Recommendation format: each recommendation has a profile applicability, description, rationale, impact, audit procedure (console and CLI), remediation, default value, references and mappings to CIS Controls safeguards. Recommendations are marked Automated or Manual depending on whether the audit can be fully scripted.
- Sections: the AWS Foundations Benchmark groups recommendations into identity and access management, storage, logging, monitoring and networking (in v5.0.0, for example, section 1 is IAM, 2.1 is S3, 3 is logging and 5 is networking). The other clouds follow a similar pattern with provider-specific sections such as databases, Kubernetes or key management.
Representative AWS requirements and their Security Hub CSPM controls (CIS AWS v5.0.0)
| CIS v5.0.0 | Requirement theme | Security Hub CSPM control |
|---|---|---|
| 1.3 | No root user access key | IAM.4 |
| 1.4 and 1.5 | MFA, and hardware MFA, on the root user | IAM.9, IAM.6 |
| 1.11 | Remove credentials unused for 45 days | IAM.22 |
| 1.13 | Rotate access keys every 90 days or less | IAM.3 |
| 1.19 | IAM Access Analyzer external access analyzer enabled | IAM.28 |
| 2.1.4 | S3 Block Public Access enabled | S3.1, S3.8 |
| 3.1 | Multi-Region CloudTrail with read and write management events | CloudTrail.1 |
| 3.7 | VPC flow logging in all VPCs | EC2.6 |
| 5.3 and 5.4 | No security group ingress from 0.0.0.0/0 or ::/0 to remote administration ports | EC2.53, EC2.54 |
| 5.7 | EC2 instances use IMDSv2 | EC2.8 |
Numbering changes between versions: AWS documents requirements CIS removed, requirements replaced by new ones, and requirements that are manual checks, such as several monitoring recommendations. Always map by version.
Identity and NHI requirements
Identity recommendations are the largest and most consequential part of each foundations benchmark, and many of them concern machine credentials rather than people.
- Root and global admin: AWS requires no root access keys and MFA (hardware MFA in Level 2) on the root user; the Azure and Google Cloud benchmarks similarly focus on protecting privileged directory and organisation roles and enforcing MFA. See the AWS IAM Policies and Roles Guide.
- Long-lived keys: AWS asks for access key rotation within 90 days and removal of credentials unused for 45 days. The Google Cloud benchmark addresses user-managed service account keys and their rotation, and service accounts holding admin roles. See the Google Cloud Service Account Keys Security Guide.
- Policy attachment and separation of duties: AWS requires policies to be attached to groups or roles rather than users (IAM.2); the Google Cloud benchmark covers which principals may act as or mint tokens for service accounts.
- External access: IAM Access Analyzer in AWS, and guest and external sharing settings on other clouds, catch identities outside your organisation.
- Metadata credentials: the AWS requirement for IMDSv2 protects the instance role credentials that workloads use, a classic NHI theft path.
Rotation rules are a floor, not the goal. A 90-day rotation requirement for access keys is satisfied by rotating; it is better satisfied by having no long-lived keys at all, using roles, workload identity federation and managed identities. See the Workload Identity Federation Guide.
Implementation steps
- Pick the target per cloud: record benchmark, version and profile in policy. Most organisations adopt Level 1 everywhere and Level 2 for regulated or production environments.
- Read the current PDF: download the latest version from CIS, note recommendations that changed, and compare against the version your scanner supports. Decide whether you assess against the scanner's version (easy evidence) or the current version (better coverage, with manual gaps).
- Baseline in audit mode: enable the CIS standard in your CSPM across all accounts, subscriptions and projects, without automatic remediation. On AWS, allow up to 18 hours for some findings to appear, and enable the newer version before disabling an older one to avoid gaps.
- Triage by risk: fix identity, logging and public exposure first (root keys, MFA, CloudTrail or equivalent, public storage, open administration ports), then the rest.
- Prevent rather than detect: enforce what you can with guardrails, such as AWS service control policies, Azure Policy deny effects or Google Cloud organisation policies, so new resources are compliant by default. See the Azure Policy and Management Groups Guide and the Google Cloud Organization Policy Guide.
- Codify: add the same checks to infrastructure-as-code scanning in pipelines so drift is caught before deployment.
- Handle exceptions: document each non-applicable or accepted recommendation with owner, reason, compensating control and review date, and suppress it in the tool with a matching note.
- Roll back safely: introduce preventive guardrails to a pilot account or folder first; if a deny policy breaks a workload, revert the guardrail through code, keep the detective check, and re-plan.
Choosing how to scan
| Option | Strength | Limitation | Fits |
|---|---|---|---|
| Native CSPM (Security Hub CSPM, Defender for Cloud, Security Command Center) | No extra infrastructure; integrated findings; some versions CIS certified | Often an older benchmark version; one cloud per tool unless connectors are licensed | Single-cloud or small teams |
| Third-party multi-cloud CSPM | One view across clouds; often faster version updates | Cost; read-only cross-account roles become high-value NHIs | Multi-cloud estates |
| Open-source scanners | Transparent checks, scriptable, cheap | Benchmark version support varies by release; you own operation | Engineering-led teams, pipelines |
| CIS-CAT Pro and Build Kits | Produced by CIS | CIS-CAT Pro's documentation lists Microsoft 365 among cloud assessments; check coverage for your cloud | SecureSuite members |
Every scanner needs read access to the whole estate. Treat scanner roles and service principals as privileged NHIs: read-only, federated rather than key-based, scoped to the organisation and monitored. See the Cloud Workload Identity Guide.
Small versus large organisations: a small team on one cloud can usually rely on the native CSPM standard at Level 1 and accept its version lag, documenting the newer recommendations as manual checks. A large or multi-cloud organisation should set one benchmark policy per cloud, assess centrally across all accounts and subscriptions, and review each new CIS release within a quarter of publication.
Evidence and assessment
- Automated results: export CSPM findings per account with the standard version, run date and pass or fail per recommendation. Keep history to show trend, not just a snapshot.
- Manual recommendations: keep screenshots, CLI output or procedure records for recommendations scanners cannot test, signed off by the control owner.
- Exceptions register: one entry per recommendation that is not applied, with rationale and compensating control.
- Certification claims: AWS states Security Hub CSPM holds CIS Security Software Certification for CIS AWS v5.0.0, v3.0.0 and v1.2.0 at Level 1 and Level 2, and Google states CIS reviews its Security Health Analytics mappings. Certification of the tool is useful evidence that checks match the benchmark, but it does not certify your environment.
Metrics
- Pass rate per benchmark and profile, per account or project, trended monthly.
- Time to remediate failed identity and public exposure recommendations (target days, not weeks).
- Percentage of accounts, subscriptions and projects enrolled in the CIS standard (target 100 percent).
- Number of open exceptions past review date.
- Gap between the version assessed and the current CIS version, in major versions.
Mapping to other frameworks
| Framework | Reference | Relationship |
|---|---|---|
| CIS Controls v8.1 | Control 4 Secure Configuration of Enterprise Assets and Software (4.1 Establish and Maintain a Secure Configuration Process) | Each benchmark recommendation maps to CIS Controls safeguards; benchmarks are the how for Control 4. See the CIS Control 4 Secure Configuration of Enterprise Assets and Software Guide. |
| NIST SP 800-53 Rev. 5 | CM-6 Configuration Settings; CM-7 Least Functionality; CA-7 Continuous Monitoring; RA-5 Vulnerability Monitoring and Scanning | Benchmarks supply the settings; CSPM scanning provides monitoring |
| NIST CSF 2.0 | PR.PS-01; ID.RA-01; DE.CM-09 | Configuration management, weakness identification and monitoring |
| ISO/IEC 27001:2022 Annex A | A.8.9 Configuration management | Benchmarks as the documented secure configuration |
| Provider frameworks | AWS, Azure and Google well-architected and security benchmarks | Overlap heavily; CIS is the provider-neutral reference. See the Microsoft Cloud Security Benchmark Guide. |
Common pitfalls
- Version confusion: a report says "CIS AWS 100 percent" against v1.2.0 while the current version adds storage, IMDSv2 and Access Analyzer requirements that were never assessed.
- Foundations only: account settings pass while compute, database and Kubernetes services, covered by separate benchmarks, are never checked.
- Exposed storage that a check would catch: the Twilio TaskRouter SDK Compromise 2020 stemmed from an S3 bucket policy that let any AWS identity write objects, left in place for years after troubleshooting; public access and bucket policy checks exist to catch exactly this drift.
- Rotation theatre: keys rotated on schedule to satisfy the check, but copies remain in CI variables and laptops, so the risk is unchanged.
- Level 2 applied blindly: recommendations with operational impact rolled out everywhere, broken workloads, and the whole programme paused.
- Manual items ignored: the dashboard shows only automated checks, and the assessor finds manual recommendations with no evidence.
Practitioner checklist
- Record the benchmark, version and profile for each cloud in your security policy.
- Compare your scanner's supported version with the current CIS version and document the gap.
- Enrol every account, subscription and project in the CIS standard.
- Prioritise identity, logging and public exposure failures.
- Replace long-lived access and service account keys instead of only rotating them.
- Turn the most important checks into preventive guardrails and pipeline checks.
- Assess the service and Kubernetes benchmarks that match the services you run.
- Keep evidence for manual recommendations and a dated exceptions register.
- Treat scanner roles as privileged NHIs: read-only, federated and monitored.
Standards and references
- CIS: CIS Benchmarks
- CIS: Amazon Web Services Benchmarks
- CIS: Google Cloud Computing Platform Benchmarks
- CIS: CIS Benchmarks FAQ
- AWS: CIS AWS Foundations Benchmark in Security Hub CSPM
- Microsoft Learn: Regulatory compliance standards in Microsoft Defender for Cloud
- Google Cloud: Compliance Manager overview
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: CIS Benchmarks Guide · CIS Controls Guide · AWS Security Hub Guide · Security Command Center Guide · Cloud Security Guide