This page goes inside a Benchmark: how recommendations are written, which platforms are covered and what the two profile levels mean. For what CIS Benchmarks are and how they differ from CIS Controls, see our CIS Benchmarks guide. Rollout and audit use are covered in implementing CIS Benchmarks and using them in audits.
Anatomy of a recommendation
A Benchmark isn't a loose list of suggestions. Every recommendation inside one, and a single Benchmark can contain several hundred, generally includes the parts below, although the exact fields and their names can vary by Benchmark version and type. That structure is what makes the guidance usable by an engineer rather than just readable by one:
- Description: what the setting is and what it controls.
- Rationale: why the change matters and what kind of attack or exposure it closes off.
- Audit procedure: the exact steps or command-line check needed to confirm whether a system currently meets the recommendation.
- Remediation procedure: the exact steps or commands needed to bring the system into compliance if the audit fails.
- Default value: what the setting is left at out of the box, before anyone changes anything.
That structure comes out of CIS WorkBench, the collaborative platform where the consensus process actually happens. Volunteer experts from vendors, government agencies, auditors and industry submit, argue over, test and vote on each recommendation before it is published, and Benchmarks are revised on an ongoing basis as platforms change and new threats emerge. It is a slower process than one vendor writing its own hardening guide, and that is the point: a recommendation that survives review from people with competing incentives is a stronger baseline than one written by whoever built the product.
Examples of CIS Benchmarks by platform
"CIS Benchmark" isn't one document, it's a family of them, each scoped to a specific platform. A few of the ones organisations reach for most often:
| Platform | Benchmark |
|---|---|
| Microsoft 365 | CIS Microsoft 365 Foundations Benchmark |
| Windows Server | CIS Microsoft Windows Server Benchmark |
| Amazon Web Services | CIS Amazon Web Services Foundations Benchmark |
| Microsoft Azure | CIS Microsoft Azure Foundations Benchmark |
| Google Cloud | CIS Google Cloud Platform Foundation Benchmark |
| Kubernetes | CIS Kubernetes Benchmark |
| Cisco network devices | CIS Cisco IOS/IOS XE Benchmark |
Each Benchmark is versioned to a specific platform release, and CIS updates or retires guides as vendors ship new ones. Check CIS's own Benchmark catalogue for the current version rather than working from a cached copy someone downloaded a year ago.
Level 1 and Level 2: two tiers of hardening
Many CIS Benchmarks organise their recommendations into Level 1 and Level 2 profiles, and some include additional profiles, such as STIG, depending on the platform. Check the documentation for the specific Benchmark version before you choose. Level 1 covers foundational hardening: settings that meaningfully reduce attack surface with minimal risk of breaking normal operation. It is intended as a baseline suitable for practically any system, including ones running general business applications where uptime matters more than defence-in-depth. Level 2 goes further, covering settings intended for higher-security environments where some reduction in functionality or performance is an acceptable trade for stronger protection. Level 2 is worth considering for regulated environments, systems handling sensitive data and high-exposure infrastructure, but only after a risk assessment and compatibility testing. Internet exposure alone does not make Level 2 the right choice.
A practical approach is to treat Level 1 as the starting point for the systems a Benchmark covers, once you have confirmed the exact platform version and that each recommendation applies. Apply Level 2 selectively, after a risk assessment, compatibility testing and approval of any exceptions, to the systems that justify the operational trade-off. The right profile is a decision made per system, not a single setting applied uniformly across the whole estate.