We are currently revamping our website to improve user experience. During this time, some pages or links may not function as expected. We regret any inconvenience and appreciate your patience and continued support.

Kubernetes Security Checklist for Production Clusters

A Kubernetes security checklist for the platform lead, IT head or security lead in India or the GCC with a production cluster on AWS, Azure or its own servers. Fourteen checks in four areas each name the record to ask for and an owner. Three marks, four checks weighted double and three plain verdicts show where the cluster stands, and a fix order shows what to do first. No tools or prices.

This Kubernetes security checklist scores one production cluster with 14 checks across access, workloads, supply chain and upkeep, each against a record the team shows. The total lands in one of three verdicts, and then a short fix order follows.


Last reviewed 2026-11-16 by Manish Sharma.

Kubernetes runs and schedules containers across a group of servers, while the scope here is one cluster in production. A record beats a belief: a setting a person can show is worth more than a setting a person remembers. The run burden behind these checks is covered in the upkeep that self-run clusters ask of a team.

In short:

  1. A check earns points only on a record the team shows, because a belief proves nothing.
  2. Exposure, admin rights, secrets and audit records weigh double, because each one opens every workload.
  3. One Absent mark on those four caps the verdict, however the rest scores.

Kubernetes Security Checklist: How Each Check Earns Its Mark

The reviewer gives every one of the 14 checks a single mark from three, based on what the team actually showed.

Mark Plain meaning Points (single / double)
Seen running The reviewer saw the setting on the live system or its output 2 / 4
Verbal only The team says yes and shows nothing 1 / 2
Absent No answer or no setting 0 / 0

Checks 1, 2, 6 and 12 count double, because each opens every workload: exposure, admin rights, secrets and audit records. The maximum is therefore 36. If a check does not apply, write it down with its reason, and its points come off the maximum. When namespaces differ, read three, including the busiest.

The bands are:

  • 30 to 36: Ready for production traffic.
  • 20 to 29: Patch named items first.
  • Under 20: Rebuild the basics first.

However, a double-weight check at Absent caps the result at the middle verdict. Also, the total is a reading on one date, not a certificate.

A record beats a belief

A check earns points only when the team shows the setting, not when it says the setting exists.

Three marks for each Kubernetes security checklist item: seen running, verbal only and absent, with four checks counted double
Four of the 14 checks count double.

Kubernetes Security Checklist, Access: Who Can Reach the API and Nodes

Three checks decide who can reach the control plane and what they can do once in.

Check Record to ask for Owner
1. Exposure (double) Yes if the team shows the API, kubelet and etcd not exposed publicly and etcd write access limited, as the checklist on kubernetes.io asks Platform lead
2. Admin rights (double) Yes if the team shows the built-in superuser group unused after setup and access reviewed at least every 24 months Security lead
3. Workload identity Yes if the team shows service account tokens mounted only in pods that need them Lead developer

On check 1, OWASP warns that write access to etcd is equivalent to root on the entire cluster, as its Kubernetes Security Cheat Sheet puts it. OWASP also says Kubelets allow unauthenticated access to their API by default, and it recommends turning on kubelet authentication and authorization. Then ask the team to show both are on.

Ask for the output

A saved command result or screenshot with a date is a record, but a promise is not.

On check 2, role-based access (role-based access) should give each person only the rights the job needs, because wider rights widen every mistake.

Kubernetes Security Checklist, Workloads: Pods, Network and Secrets

Three checks cover what runs inside Kubernetes and how it talks, before the supply chain checks.

Check Record to ask for Owner
4. Pod rules Yes if the team shows Pod Security Standards enforced on every namespace and memory limits set Platform lead
5. Network Yes if the team shows a network plugin that enforces policies, default deny per namespace and a filtered metadata endpoint Network lead
6. Secrets (double) Yes if the team shows Secrets encrypted at rest in etcd and none in ConfigMaps Security lead

The Pod Security Standards define three policies that cover the security spectrum, from open to restricted. Pick one per namespace, then enforce it. We read all three sources on 16 November 2026, but none shows a page date.

For check 5, the checklist asks the team to filter access from workloads to the cloud metadata API. Also ask the network lead to show the rule, not describe it.

Default network rules

OWASP says Kubelets allow unauthenticated access to their API by default, so ask for proof that authentication is on.

Kubernetes Security Checklist, Supply Chain: Images and Registries

Three checks cover what gets into Kubernetes, while one sentence applies to every image in it.

Check Record to ask for Owner
7. Image identity Yes if the team shows images pinned by digest, or a signature checked at deploy Platform lead
8. Image scanning Yes if the team shows scans at build and in deployment, with known flaws patched Security lead
9. Registry access Yes if the team shows pushing limited to named people Platform lead

For check 7, the Images section of the kubernetes.io checklist says an admission controller can verify image signatures automatically at deploy time.

Container Security Checklist Items for Each Image

A container security checklist for one image is short. Run as an unprivileged user, start from a small base, and rebuild the image, then redeploy it, instead of patching it in place. For the build itself, see the build steps for a small image.

Upkeep Checks for Production Clusters

Three more checks keep Kubernetes safe after day one, because systems drift.

Nodes, Versions and Backups

Check 10 asks for node patching by rebuild and redeploy: the team shows the last rebuild date, and also the image it used. OWASP’s cheat sheet says to update the source image, then redeploy, and not patch running containers.

Check 11 asks whether Kubernetes runs a supported version. Support windows are a separate topic, so this article gives no dates.

Check 14 asks whether the team backs up cluster state and has tried a restore. A backup stays Verbal only until someone has restored it.

Audit Logs and Runtime Alerts

Check 12 (double) asks for audit logs sent to separate storage, with access limited, Secrets logged at metadata level only and failed authorization reviewed. OWASP says to log secrets and configmaps at metadata level only.

Check 13 asks whether runtime alerts reach a named person, because an unowned alert goes unread. An alert that lands in a shared inbox with no owner is Verbal only at best.

Kubernetes Security Checklist Scored on a Sample Cluster

The example below is illustrative and runs an order service, but it is not a Signisys project.

Check Mark Points
1 Exposure Seen running 4
2 Admin rights Verbal only 2
3 Workload identity Verbal only 1
4 Pod rules Seen running 2
5 Network Verbal only 1
6 Secrets Seen running 4
7 Image identity Absent 0
8 Image scanning Seen running 2
9 Registry access Seen running 2
10 Node patching Seen running 2
11 Supported version Seen running 2
12 Audit logs Verbal only 2
13 Runtime alerts Absent 0
14 Backups Verbal only 1

The total is 25 of 36, so the verdict is Patch named items first. The named items are checks 2 and 12, then 7 and 13.

Key Takeaway The sample scores 25 of 36, the middle verdict, so the double-weight checks set the fix order.

What to Fix First on a Production Cluster

Signisys’s working view is to fix the double-weight checks that sit below Seen running first, then the network check, then the supply chain, then upkeep. Then score again after each batch. Our reasoning is that the four double-weight checks each open every workload, while the others limit the damage after that.

On the sample cluster, that order starts with checks 2 and 12, then moves on to 7 and 13. If a gap needs a tool, security tool categories sit in the kinds of tool that watch workloads.

Fix order for a production cluster: double-weight checks first, then network, supply chain and upkeep
Score again after each batch.

Key Takeaway Fix the four double-weight checks first, then score again after each batch, because each batch changes the total.

Where These Checks End

Who runs which layer when a provider runs the control plane is set out in a comparison of the two ways to run a cluster.

Also, checks on the cloud account sit in checks on the cloud account around those workloads. Functions, however, have their own list of readiness marks for functions.

Signisys provides these services: see our review for EKS and AKS clusters, our modernization and observability solution and DevOps and platform automation services. A working session scores one cluster with your own records, with no promised result and no price. No charge, with a fixed scope.

Book a consultation
Score your production cluster with us

Kubernetes Security Checklist FAQ

Should a staging cluster get the same 14 checks?

Yes, the same 14, scored lighter on checks 12 and 13. Production, however, gets the full read.

How do Kubernetes security best practices relate to these 14 checks?

Practices are advice, while each check turns one into a record to show. The sources are kubernetes.io and OWASP.

Can one engineer score a cluster alone?

Yes, but Signisys’s working view is that a second person should read the Seen running marks. Two readers catch settings that one reader assumed, because the second tests each mark.

What if each namespace has a different owner?

Score checks 3 to 5 per namespace, then take the lowest mark into the overall score.

Newsletter

Security Insights, Delivered.

Practical, vendor-neutral analysis of threats, tools, and architecture decisions.

Book a Consultation

A practice principal will contact you within 1 business day.

Thank you! A Signisys principal will be in touch within 1 business day.