A Kubernetes security assessment checks how your clusters are built, configured and run: the control plane and nodes, who can do what through access control, how workloads are isolated, how network traffic and secrets are handled, what goes into your images, and whether you would detect an attack. Good assessments combine a configuration review against recognised benchmarks with hands-on testing.
Kubernetes gives engineering teams a lot of power and a lot of settings. Many defaults favour getting things running over keeping them locked down, and a cluster can drift a long way from its original design as teams add workloads. An assessment tells you where you stand and what to fix first. This article explains what is in scope, the standards assessors use and how the work is done.
Executive Summary & Key Takeaways
- Several Kubernetes defaults are permissive. Pods can talk to each other freely and Secrets are stored unencrypted unless you configure otherwise.
- Access control is usually where the most serious findings sit, because a few powerful permissions can give control of a whole cluster.
- Assessors work from public benchmarks, including the CIS Kubernetes Benchmark, the NSA and CISA Kubernetes Hardening Guide and the Kubernetes Pod Security Standards.
- Managed services such as EKS, AKS and GKE take the control plane off your hands, but your workloads, permissions and network policies are still yours to secure.
- A configuration review finds misconfiguration. Hands-on testing shows how far an attacker could get from a compromised pod or credential.
Why Kubernetes needs its own assessment
A general cloud or infrastructure test will not look closely at how Kubernetes itself is set up. Some of the defaults are worth knowing:
- Network traffic is open by default. If no network policies exist in a namespace, all traffic to and from its pods is allowed. Policies also only work if the cluster’s network plugin enforces them (Kubernetes documentation, retrieved 10 November 2026).
- Secrets are not encrypted by default. Kubernetes Secrets are stored unencrypted in etcd unless encryption at rest is configured. Anyone allowed to create a pod in a namespace can use that to read any Secret in that namespace (Kubernetes documentation, retrieved 10 November 2026).
- Some groups bypass access control entirely. Members of the
system:mastersgroup bypass all role-based access control checks, and removing role bindings does not revoke that access (Kubernetes documentation, retrieved 10 November 2026).
None of these are flaws in Kubernetes. They are design choices that put security decisions in your hands, and an assessment checks whether those decisions have been made well.
What a Kubernetes security assessment covers
Cluster configuration
Assessors check the configuration of the control plane (the API server, scheduler, controller manager and etcd) and of the worker nodes, including the kubelet. The CIS Kubernetes Benchmark is the usual reference, with sections for control plane components, worker nodes and policies (CIS, retrieved 10 November 2026). They also check you are running a supported version. The Kubernetes project maintains the three most recent minor releases, each with about a year of patch support (Kubernetes, retrieved 10 November 2026), so clusters that are not upgraded regularly fall out of support quickly.
Identity and access control
Role-based access control (RBAC) decides what users, pipelines and workloads can do. Assessors look for wide use of cluster-admin, wildcard permissions, and rights that allow privilege escalation, such as permission to create workloads, read Secrets, or use the escalate, bind and impersonate verbs (Kubernetes documentation, retrieved 10 November 2026). They also check how service account tokens are issued and whether workloads have more cloud permissions than they need, since a compromised pod with powerful cloud credentials can reach well beyond the cluster.
Workload isolation
This covers what your containers are allowed to do on the node. Kubernetes defines three Pod Security Standards: Privileged, which is unrestricted; Baseline, which prevents known privilege escalations; and Restricted, which follows current hardening practice (Kubernetes documentation, retrieved 10 November 2026). Assessors check which standard each namespace enforces, and look for privileged containers, access to the host’s network or filesystem, and containers running as root.
Network segmentation
Assessors map which workloads can reach which, and compare that with what should be allowed. They check that network policies exist, that they are enforced, and that sensitive services such as databases and internal admin tools are not reachable from every pod.
Secrets management
This covers how credentials, keys and certificates reach your workloads: whether encryption at rest is enabled, who can read Secrets, whether an external secrets manager is used, and whether secrets have leaked into images, environment variables or configuration files.
Images and the software supply chain
Most of what runs in a cluster comes from container images and the pipeline that builds them. Assessors look at image sources, vulnerability scanning, whether only approved images can run, and who can push to your registries. This is where cluster security meets CI/CD security.
Logging and detection
Finally, assessors check whether you would notice an attack: whether Kubernetes audit logging is enabled, where logs go and whether anyone would be alerted to suspicious activity. The NSA and CISA guidance lists capturing and monitoring audit logs among its core mitigations (NSA and CISA, retrieved 10 November 2026).
The standards assessors use
| model | Published by | What it is for |
|---|---|---|
| CIS Kubernetes Benchmark | Center for Internet Security | Detailed configuration checks; separate versions exist for EKS, AKS, GKE and other platforms |
| Kubernetes Hardening Guide (version 1.2, August 2022) | NSA and CISA | Hardening guidance across pods, networks, authentication, logging and upgrades |
| Pod Security Standards | Kubernetes project | Three policy levels for what workloads may do |
| OWASP Kubernetes Top 10 | OWASP | A prioritised list of common Kubernetes risks |
The OWASP Kubernetes Top 10 published a 2025 list, which adds “cluster to cloud lateral movement” as a named risk, and the project is still inviting feedback on it (OWASP, retrieved 10 November 2026).
How the assessment is done
Most assessments combine two kinds of work.
A configuration review checks the cluster against the benchmarks above, using read-only access, configuration exports and your manifests or infrastructure as code. It is thorough and repeatable, and it finds misconfiguration at scale.
Hands-on testing starts from realistic positions, such as a compromised application pod, a leaked developer credential or an exposed service, and tries to see how far an attacker could get: into other namespaces, onto the node, to Secrets, or out to your cloud account. This shows which findings actually matter and how they chain together. It is closer to a penetration test and follows the same scoping and authorisation process.
For an example of this work in practice, see our Kubernetes platform security testing case study.
Managed or self-managed clusters
On a managed service, the provider runs the control plane, so some benchmark checks do not apply to you. That is why CIS publishes separate benchmarks for managed platforms (CIS, retrieved 10 November 2026). The rest of the assessment is largely the same: access control, workloads, network policies, secrets, images and logging are still your responsibility. Self-managed clusters add the control plane, etcd and node operating systems to the scope.
Multi-tenant clusters
If several teams or customers share a cluster, isolation between them becomes the main question. The Kubernetes documentation advises using namespaces to separate workloads that need different levels of trust, while noting that boundaries within a namespace should be considered weak (Kubernetes documentation, retrieved 10 November 2026). If you run a SaaS platform, see our article on tenant isolation testing.
Common mistakes
- Assuming a managed service is secure by default. The provider secures the control plane; your workloads and permissions are still yours.
- Granting
cluster-adminto pipelines and developers for convenience, then never removing it. - Writing network policies that nothing enforces because the cluster’s network plugin does not support them.
- Scanning images but not checking what runs. A clean registry does not help if any image can be deployed.
- Letting clusters fall behind on versions, so security fixes stop arriving.
Frequently asked questions
Do we need an assessment if we use EKS, AKS or GKE?
Yes, though the scope is smaller. The provider handles the control plane, but most serious findings in Kubernetes come from access control, workload settings and network policy, which remain your responsibility on every managed service.
How is this different from a container image scan?
An image scan looks for known vulnerabilities in the software inside your images. An assessment looks at the whole cluster: who can deploy what, what containers are allowed to do, how they talk to each other and how far an attacker could get. Image scanning is one input to an assessment, not a replacement for it.
When is the right time for a Kubernetes assessment?
Before a cluster goes into production, after significant changes such as a new platform design or a move to multi-tenancy, and ahead of audits or large customer security reviews. Between assessments, continuous posture monitoring catches drift.
What access do assessors need?
Usually read-only access to the cluster and the cloud account, plus manifests or infrastructure-as-code repositories for the configuration review. Hands-on testing may need a test workload or a test account at an agreed privilege level. All of this is agreed in writing before work starts.
Getting started
A Kubernetes assessment gives you a clear picture of how your clusters are configured, how far an attacker could get and what to fix first. Our cloud security service then keeps clusters configured securely over time, including Kubernetes access control and admission policies.
To scope an assessment of your clusters, talk to our team.

Jared Henderson
Writing about cyber security and technology at Falx.





