CIS Benchmarks are the configuration standard most auditors, customers and cloud marketplaces recognise, yet many teams adopt them as an all-or-nothing scan score rather than as a tailored, evidenced baseline. The Center for Internet Security (CIS) publishes more than 100 vendor-neutral benchmarks for operating systems, cloud platforms, server software, network devices, mobile devices and DevSecOps tools, each built by consensus and organised into profiles such as Level 1, Level 2 and STIG. This guide explains how a benchmark is structured, how to choose a profile, how to assess with CIS-CAT and other SCAP tools, and how to run an exception process that stands up in an audit. It goes deeper than the CIS Control 4 Secure Configuration of Enterprise Assets and Software Guide, which covers the wider secure configuration process, and is written for platform engineers, compliance leads and security architects.
Key takeaways
- Level 1 is the default, Level 2 is a decision. CIS describes Level 1 as a base set that can be implemented fairly promptly while keeping systems usable, and Level 2 as defence in depth that can cause problems if applied without care.
- Profiles are role-specific. A Windows Server benchmark distinguishes domain controllers from member servers, and Linux benchmarks distinguish servers from workstations, so pick the profile that matches the asset's role.
- A scan score is not compliance. Manual recommendations and documented exceptions both affect the result; the evidence is the score plus the tailoring record.
- Exceptions need structure. Every recommendation you do not apply needs a reason, a compensating control, an approver and an expiry, and should be encoded in the assessment profile so it does not reappear as a failure.
- Many recommendations are identity controls. Password and lockout policy, privileged group membership, SSH and sudo settings, and service account rights are a large share of every OS benchmark.
What it is and who it applies to
A CIS Benchmark is a prescriptive configuration guide for one product and version, for example a Windows Server release, a Linux distribution, Kubernetes, a cloud provider account or a database. CIS lists benchmarks across cloud providers, desktop software, DevSecOps tools, DISA STIG aligned content, mobile devices, network devices, operating systems, server software and storage devices.
- Development: benchmarks are produced through a consensus process involving volunteers, vendors and CIS staff. Drafts are discussed on CIS WorkBench until consensus is reached, and reported discrepancies feed later releases. Release timing follows the community and the vendor's schedule; CIS says its Windows benchmarks cover the latest Windows build within 90 days of release.
- Who uses them: no law mandates CIS Benchmarks in general, but they are a common answer to configuration requirements in frameworks and contracts, and NIST's National Checklist Program (SP 800-70 Rev. 5, May 2026) describes the role of such checklists: configuring a product for a risk level, verifying the configuration, detecting unauthorised changes and producing evidence.
- Access and licensing: CIS states the benchmarks are free to download in PDF format for non-commercial use. Word and XML formats, CIS-CAT Pro, build kits and tailoring in WorkBench require CIS SecureSuite membership. Check the licence terms before embedding benchmark text or content in a commercial product or service.
- Related products: build kits (Group Policy objects for Windows, shell scripts for Linux and Unix) and CIS Hardened Images (preconfigured virtual machine images on major cloud marketplaces) implement benchmark settings for you.
Structure and key requirements
Profiles
| Profile | CIS description | When to use | Trade-off |
|---|---|---|---|
| Level 1 | A base recommendation set that can be implemented fairly promptly, reducing attack surface while keeping machines usable | Default for all in-scope assets | Leaves some defence in depth measures out |
| Level 2 | Defence in depth for environments where security is paramount; extends Level 1 and is not standalone | High-value servers, regulated or internet-facing systems, privileged access workstations | Higher chance of breaking applications or support tooling |
| STIG | Replaces the former Level 3; includes STIG-specific recommendations plus overlapping Level 1 and Level 2 items | Organisations that must show DISA STIG alignment but want CIS tooling | Strictest; tailor carefully. See the DISA STIG Guide |
| Product-specific extras | Some benchmarks add profiles, for example Windows client benchmarks include optional BitLocker and Next Generation Windows Security profiles | Where the feature is licensed and supported on your hardware | Hardware and licence dependencies |
Profiles are also split by role. Windows Server benchmarks have separate Level 1 and Level 2 profiles for domain controllers and member servers, and some benchmarks have stand-alone variants for non-domain systems. Linux benchmarks typically have Server and Workstation variants of each level.
Anatomy of a recommendation
- Title and status: each recommendation is numbered within its section and marked Automated (a tool can check it) or Manual (it needs human review).
- Profile applicability: the profiles it belongs to. Every recommendation is tied to at least one profile.
- Description, rationale and impact: what the setting does, why it matters and what may break. Read impact before rolling out, not after.
- Audit and remediation: how to check the setting and how to set it, often with the exact Group Policy path, registry value, configuration file line or CLI command.
- Default value and references: the vendor default and supporting documentation.
- CIS Controls mapping: the CIS Controls safeguard and Implementation Group the recommendation supports.
Operating system benchmarks are organised by area. Windows benchmarks follow the Group Policy tree (account policies, local policies including user rights assignment and security options, system services, Defender Firewall, advanced audit policy, then administrative templates). Linux benchmarks cover initial setup (filesystems, bootloader, mandatory access control), services, network parameters, host firewall, access control (SSH, sudo, PAM, password policy), logging and auditing, and system maintenance such as file permissions.
Identity and NHI requirements
A large part of every operating system benchmark controls who can authenticate and with what privileges. These are the recommendations to review first, because they often conflict with how service accounts and automation actually work.
| Area | Typical benchmark requirement | Non-human identity impact |
|---|---|---|
| Password and lockout policy | Minimum length, history, lockout threshold and duration | Service accounts with static passwords hit lockout or expiry; move them to managed accounts. See the Group Managed Service Accounts (gMSA) Guide |
| User rights assignment (Windows) | Restrict "Log on as a service", "Log on as a batch job", "Access this computer from the network" and deny rights for local accounts | Scheduled tasks and agents fail unless their accounts are listed explicitly; this is also where over-privileged service accounts surface |
| Local administrator accounts | Rename or disable the built-in administrator, restrict local accounts from network logon | Pair with unique, rotated local admin passwords. See the Local Administrator Password Solution (LAPS) Guide |
| SSH server (Linux) | Disable root login, restrict allowed users or groups, limit authentication methods and ciphers | Automation that logs in as root or uses unmanaged keys breaks or must be redesigned. See the SSH Key Management Guide |
| sudo and PAM | Require sudo logging and re-authentication, configure password quality and faillock | Passwordless sudo for deployment accounts is a common exception; scope it to specific commands |
| System accounts (Linux) | System accounts have non-login shells; no accounts with empty passwords | Application accounts created by installers often violate this |
| Cloud foundations benchmarks | Root or owner account MFA, no root access keys, key rotation, least privilege roles | Directly governs access keys and workload roles. See the CIS Cloud Foundations Benchmarks Guide |
Treat each failure in these areas as an identity finding as well as a configuration one: record which account is affected, its owner and whether it is human or non-human. See the CIS Control 5 Account Management Guide.
Implementation steps
- Inventory and scope: list each product and version in the estate and map it to the matching benchmark version. Unsupported versions only have archived benchmarks, which is itself a finding.
- Choose profiles per asset class: Level 1 for the general estate, Level 2 for defined high-value classes, STIG profile only where required. Write the choice into your configuration standard.
- Baseline scan before changing anything: assess a representative sample to see current conformance and which recommendations fail most often.
- Review impact statements: for every failing recommendation, read the impact section and consult application owners. Decide adopt, adopt with a different value, or exception.
- Build the tailored baseline: encode decisions in a tailored profile (WorkBench for CIS-CAT, or a tailoring file for SCAP tools) and in your build artefacts (GPOs, Intune profiles, Ansible roles, image pipelines). Build kits and hardened images can be the starting point.
- Pilot in rings: apply to test systems, then a pilot group, then production by role. For Windows, use separate GPOs per profile so rollback is unlinking an object; for Linux, keep changes in configuration management so a revert is a commit.
- Bake it in: apply the baseline at build time (golden images, templates, infrastructure as code) rather than only remediating running systems.
- Assess continuously: schedule assessments, feed results to a dashboard, and track drift. See the Configuration Drift and Baseline Enforcement Guide.
- Update on release: when CIS publishes a new benchmark version, compare it with your tailored profile and re-review only the changed recommendations.
Choosing between implementation routes
| Route | Good for | Limitations |
|---|---|---|
| CIS Hardened Images | Cloud workloads where you can start from a marketplace image | Still needs your tailoring and patching; cost per instance |
| Build kits (GPO, scripts) | Fast initial rollout on Windows domains and Linux fleets | Member-only; must be merged with your own policies |
| Configuration management (Ansible, DSC, Intune, Puppet) | Large or mixed estates needing drift control | You maintain the mapping from recommendation to code |
| Vendor baselines plus CIS delta | Windows estates already on Microsoft baselines | Requires a clear record of which source owns each setting. See the Microsoft Security Baselines and Security Compliance Toolkit Guide |
Small organisations usually get most value from Level 1 via a hardened image or build kit plus CIS-CAT Lite spot checks. Large organisations need tailored profiles per asset class, automation, and an exception workflow integrated with change and risk management.
Evidence and assessment
- CIS-CAT Pro Assessor: automates assessment against benchmark content, runs local or remote sessions against multiple targets, produces detailed HTML reports and maps results to the CIS Controls and Implementation Groups. It can assess tailored benchmarks exported from WorkBench. It requires SecureSuite membership.
- CIS-CAT Lite: free, assesses a selection of benchmarks and gives a compliance score. Useful for small estates or proving a concept, not for fleet evidence.
- Other SCAP and agent tools: many vulnerability and posture tools ship CIS-aligned policies; check that they state which benchmark version and profile they implement. See the SCAP and Security Content Automation Guide and the OpenSCAP and Linux Compliance Scanning Guide.
- Remote assessment credentials: credentialed scans use an account with read access to configuration, often administrative on Windows and sudo on Linux. That scanner account is a privileged non-human identity: use a dedicated account, vault its credential, restrict where it can log on from and rotate it.
What an assessor expects to see
- The configuration standard naming the benchmark, version and profile for each asset class.
- The tailored profile and the exception register with reason, compensating control, approver and expiry for each deviation.
- Recent assessment results per asset class, with trend over time, and evidence that Manual recommendations were reviewed.
- Remediation tickets for failures, and change records linking baseline changes to approvals.
| Metric | Healthy signal |
|---|---|
| Conformance score per asset class against the tailored profile | Above an agreed threshold (for example 95 percent) and stable |
| Assets assessed in the last 30 days | All in-scope assets |
| Open exceptions past expiry | Zero |
| Time from new benchmark release to tailored profile update | Within an agreed window, for example one quarter |
| Identity-related failures (rights assignment, SSH, local admin) | Falling, each with a named owner |
Mapping to other frameworks
| Framework | Reference | How a CIS Benchmark supports it |
|---|---|---|
| CIS Controls v8.1 | Safeguard 4.1 Establish and Maintain a Secure Configuration Process; 4.2 for network infrastructure; 4.7 Manage Default Accounts; 4.8 Uninstall or Disable Unnecessary Services | Each recommendation states the safeguard it supports |
| NIST SP 800-53 Rev. 5 | CM-2 Baseline Configuration; CM-6 Configuration Settings and CM-6(1); CA-7 Continuous Monitoring | The benchmark is the documented configuration setting source; deviations are approved and recorded |
| NIST CSF 2.0 | PR.PS-01 Configuration management practices are established and applied | Evidence of established configuration practice |
| NIST SP 800-70 Rev. 5 | National Checklist Program | Benchmarks are security configuration checklists in the sense the programme describes |
| ISO/IEC 27001:2022 | Annex A 8.9 Configuration management | Defined, documented and monitored security configurations |
| DISA STIGs | STIG profile in CIS benchmarks | Overlap is large but not identical; prove STIG compliance against the STIG itself where it is contractually required |
Common pitfalls
- Scanning against the wrong profile: domain controllers scored against member server content, or workstations against server content, produce false failures that erode trust in the results.
- Chasing 100 percent: teams apply Level 2 everywhere to raise the score, break support tooling, then roll back the whole baseline. Tailor instead.
- Exceptions held outside the tool: deviations live in a spreadsheet but the scanner still reports them as failures, so real regressions are lost in noise.
- Stale benchmark versions: the scanner runs content for an older OS release, so new settings are never checked.
- Ignoring Manual recommendations: they are not scored automatically, so they are never reviewed, although several cover account and access reviews.
- Running the scanner with a shared admin account: the credentialed scanning identity becomes a widely trusted, rarely rotated credential with network logon rights on every server.
Practitioner checklist
- Map every product and version in scope to its current CIS Benchmark and record the version in your configuration standard.
- Assign a profile (Level 1, Level 2, STIG, role variant) to each asset class and document why.
- Read the impact section for each failing recommendation before remediation and agree changes with application owners.
- Encode tailoring in the assessment tool so documented exceptions do not appear as failures.
- Keep an exception register with reason, compensating control, approver and expiry date.
- Apply the baseline at build time through images, build kits or configuration management, and roll out in rings with a tested rollback.
- Review identity-related recommendations with the owners of service accounts, SSH automation and scanning accounts.
- Use a dedicated, vaulted and rotated account for credentialed scans.
- Assess all in-scope assets at least monthly and report conformance, exception age and version lag.
- Re-review changed recommendations whenever CIS publishes a new benchmark version.
Standards and references
- CIS: CIS Benchmarks
- CIS: CIS Benchmarks FAQ
- CIS: CIS-CAT Pro
- CIS: CIS-CAT Pro Assessor v4 documentation
- CIS: Microsoft Windows Server Benchmarks
- CIS: CIS Controls v8.1
- NIST: SP 800-70 Rev. 5 National Checklist Program for IT Products
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: CIS Control 4 Secure Configuration of Enterprise Assets and Software Guide · CIS Controls Guide · Linux Server Hardening Guide · Windows Server Hardening Guide · DISA STIG Guide