Patch Management Policy

Patch management policy for a SaaS company

A SaaS company patches by rebuilding. This variant counts a container window as closed only when the rebuilt image runs in production and the old one does not, assigns managed-service patches to the provider or to you, and lists the evidence an auditor asks for. SOC 2 criteria text could not be read on 19 Nov 2026.

SaaS company variant: paste over the same-numbered sections of the general policy

1 Purpose and scope

1.1 This policy covers production and non-production workloads of [Company name]: host operating systems, container base images, application images, libraries and packages, build tooling, managed databases and queues, and employee endpoints.

1.2 For each managed service, the inventory records which patches the provider applies and which the company applies (the shared responsibility line). Provider-applied patches are monitored through [provider notices], not assumed.

3 Asset inventory

3.1 The inventory adds: image name and digest; base image and its source; package manifest or software bill of materials for each release; the pipeline that builds it; runtime location (cluster, region, account).

3.2 Inventory data is read from the cloud provider and the registry by API at least [daily].

4 Severity and patch windows

4.1 Windows follow general policy section 4.3.

4.2 For a container or image, the window ends when a rebuilt image is running in production and the vulnerable image is no longer running, not when the rebuild is merged.

4.3 Third-party and open source components are patched by updating the dependency and rebuilding. The pipeline fails a release that carries a class 1 or class 2 finding unless an exception is recorded.

5 Testing and staging

5.1 Patched images go through the same automated tests as any release. They deploy to [canary percent] of production first, then expand if error rate and latency stay within [thresholds] for [period].

5.2 Rollback is redeployment of the previous image digest.

6 Maintenance windows

6.1 Rolling replacement is the default and needs no customer-visible window. Hosts that cannot be replaced use the standard windows in general policy section 6.

6.2 Live kernel patching, where used, covers only the vulnerabilities its vendor supports. It does not replace the next scheduled reboot or replacement.

8 Reporting and metrics

8.1 Reports add: share of running workloads on an image younger than [30] days; oldest running image age; findings past window by team.

8.2 For each monthly report, export from [system]: the inventory snapshot, the list of changes with test evidence, and the exceptions list. These are the records given to an auditor.

What changes from the general policy

NIST SP 800-40 Rev. 4 recommends running applications on platforms where patching is built in, with containers as its example, and using toolchains that already rebuild applications with updated components and test them before release. A SaaS policy written around that advice differs from the general one in four places: what counts as an asset, when a window closes, who patches managed services, and what evidence is exported.

  • Asset. An image digest and its package manifest replace a hostname. NIST's own inventory list includes the party who administers the asset, with shared responsibility as an explicit value. CISA describes an SBOM as a nested inventory, a list of ingredients that make up software components (CISA).
  • Window closure. A merged fix is not a patched workload. Section 4.2 closes the clock when the old image has stopped running.
  • Managed services. NIST lists moving to a service where the provider patches as a way to transfer risk, not to remove the duty to know the state. Section 1.2 requires a record of what the provider covers.
  • Components. NIST SP 800-218 PW.4.4 asks organizations to verify that acquired commercial, open-source and other third-party components comply with the organization's own requirements through their life cycles. Section 4.3 puts that in the pipeline. The third-party and open source guide covers the practice.

Live kernel patching: what to write and what not to

If the variant allows live patching, state its limits. Red Hat says live kernel patches are for selected Important and Critical CVEs, are cumulative, are supported without extra cost for 6 months after the kernel errata release date, and exist for x86_64 and ppc64le only (Red Hat, updated 1 September 2026). Canonical says Livepatch covers critical and high severity kernel vulnerabilities, between security maintenance windows, with coverage of 10 years under Ubuntu Pro (Canonical). Both statements are the vendors' own. Section 6.2 therefore keeps the scheduled reboot or replacement as the real fix.

Frameworks this variant touches

Clauses read 19 Nov 2026. A reasonable-time rule carries no day count; the general policy's windows are the site's suggested defaults.
SourceClauseWhere the variant answers itRead status
SOC 2 Trust Services CriteriaCC7.1, CC8.1Section 8.2 evidence exportText could not be read on 19 Nov 2026 (AICPA download needs an account). No claim is made about what these criteria require
NIS2 Implementing Regulation 2024/2690Point 6.6.1(a) to (d)Sections 4 and 5, 4.3Read. Applies to cloud computing providers among others: patches applied within a reasonable time, tested before production, sources verified
NIST SP 800-218PW.4.4, RV.1.1Section 4.3Read
NIST SP 800-40 Rev. 4Sections 3.2, 3.5.1Sections 3.1, 5.1Read

Next steps

Pair this text with the general policy sections 2, 7 and 9, and read patching containers and images for the steps behind clause 4.2.

Sources

Last reviewed 19 Nov 2026