Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
Kubernetes security covers the control plane, the nodes, the workloads, and the pipeline that feeds them. Most real incidents start with an exposed API, a weak role, or an over-privileged pod.

What Is Kubernetes Security?
Kubernetes security is the set of controls that keep a cluster, and everything it schedules, from being turned against you. It spans four layers that the community usually calls the four Cs: the cloud or data centre the cluster sits in, the cluster itself, the containers it runs, and the code inside them. A weakness at any layer tends to collapse the ones below it, which is why a single exposed dashboard has repeatedly turned into full cloud account compromise.
Every meaningful decision in a cluster passes through the API server, and the API server trusts etcd. An attacker who reaches either one owns every namespace, every secret, and every node. Three exposures account for most of what gets found in assessments:
--anonymous-auth=true, the kubelet will happily list pods and run commands inside them, bypassing the API server and its audit log entirely.The 2018 incident in which cryptomining workloads were deployed into a Tesla cluster began with a Kubernetes console that required no password, and ended with the attackers reading cloud credentials stored inside the cluster. Siloscape, documented in 2021, took the reverse route: it escaped a Windows container, then used the node credentials to deploy further workloads. Both are the same story told from different ends.
Kubernetes authorises almost everything through role-based access control, and the default posture of many clusters is far looser than teams assume. Service accounts receive automatically mounted tokens, CI systems are handed cluster-admin because it was quicker, and roles accumulate verbs nobody removes. When reviewing, look for wildcard rules, escalate and bind verbs, secret list access, and the right to create pods in a namespace that holds sensitive secrets — the last of these is effectively secret read access, because a pod can mount anything in its own namespace.
The same discipline applies to the identity the cluster presents outwards. Giving every node a broad cloud role means every pod inherits it. Workload identity binds a specific service account to a specific cloud role instead, which is the single most effective way to stop a compromised pod from reading your object storage.
PodSecurityPolicy was deprecated in Kubernetes 1.21 and removed in 1.25. Its replacement, Pod Security Admission, applies the Pod Security Standards through namespace labels in enforce, audit, or warn modes. Enforcing the restricted profile blocks privileged containers, host namespace sharing, dangerous volume types, and privilege escalation in one step. Where you need finer rules — registry allow lists, required labels, banned image tags — Kyverno, OPA Gatekeeper, or the built-in Validating Admission Policy fill the gap.
Networking deserves equal attention. Without a NetworkPolicy, every pod can talk to every other pod and to the cloud metadata endpoint, so a single vulnerable web application becomes a route to credentials. Start with a default-deny policy per namespace and open only the flows that are needed.
Audit logging is off or minimal in many clusters. A sensible policy records Metadata for secrets and RequestResponse for RBAC changes, and ships the result somewhere outside the cluster. Runtime tools such as Falco or eBPF-based agents flag the behaviour that configuration alone cannot prevent: a shell spawning in a production container, an unexpected outbound connection, or a process reading /var/run/secrets. Pair that with image provenance checks at admission so only signed, scanned images ever reach a node.
The CIS Kubernetes Benchmark gives testable settings, and kube-bench automates the check. The NSA and CISA Kubernetes Hardening Guidance, first published in August 2021 and revised in March 2022, is a readable summary of the same ground with threat context. NCSC Cloud Security Principles apply directly to a managed cluster, and for operators within scope of the NIS Regulations a cluster running essential services sits squarely inside their risk management duties.
Benchmarks measure configuration; they do not measure consequence. At Privia Security, we test clusters by starting from a compromised pod and seeing how far it reaches, as part of our cloud security services.
Managed services such as EKS, AKS, and GKE secure and patch the control plane for you, which removes a genuine burden. Everything above it remains yours: RBAC, admission policy, network policy, workload configuration, image provenance, and logging. Default settings are deliberately permissive so that workloads run on day one, so a new managed cluster typically fails a CIS Benchmark check in several places.
Close external exposure of the API server, dashboard, and kubelet, then remove cluster-admin from anything that does not need it. Next, label namespaces for the restricted Pod Security Standard in audit mode to see what would break, enable audit logging, and apply default-deny network policies. Those four changes remove the attack paths that appear most often in real compromises.
No. A NetworkPolicy controls pod-to-pod and pod-to-endpoint traffic inside the cluster and is enforced by the CNI plugin, so it is a segmentation control rather than an inspecting firewall. It has no application-layer awareness and does not log by default. Use it alongside cloud security groups, egress filtering, and a web application firewall in front of anything internet-facing.
At least annually, and after any significant architectural change such as adding multi-tenancy, a service mesh, or a new cloud integration. Continuous scanning catches configuration drift between tests, but it will not chain a vulnerable application, a permissive service account, and a reachable metadata endpoint into a single attack path. That is what a hands-on assessment is for.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.