Techbanq

Cybersecurity · Kubernetes

A default Kubernetes cluster trusts everything inside it

Out of the box, every pod can reach every pod, service accounts are mounted whether workloads need them or not, and the credentials handed out during setup quietly keep cluster-admin. Hardening is the deliberate walk back from those defaults — done in an order that protects the cluster without breaking the teams shipping on it.

Serving Saudi Arabia from our Riyadh office and our engineering hubs in Lahore and Karachi

Structured network cabling running into a switch cabinet
  1. 01

    Market

    Saudi Arabia

  2. 02

    Written for

    CISO, CTO

  3. 03

    Decision

    Harden a Kubernetes cluster beyond its defaults

  4. 04

    Languages

    EN · AR

Cybersecurity credentials

  • National Cybersecurity Authority, Saudi Arabia

    NCA registration

    Registered cybersecurity service provider

  • ISO

    ISO/IEC 27001

    ISO/IEC 27001 certified

Certifications & registrations

The hardening spine

Four defaults a cluster has to unlearn, in order

Each stage reverses one permissive default. The order matters: narrowing access first means every later policy is written and changed by people whose own privilege is already accounted for.

  1. 01

    Access narrowed

    RBAC rebuilt around what each team actually does: narrow roles bound to groups, no personal standing cluster-admin, and the emergency path recorded when it is used.

    • RBAC
    • Groups
  2. 02

    Traffic declared

    Network policies turn the flat pod network into segments: each namespace states which traffic it accepts and sends, and everything undeclared is denied — after an observe period proves the map is right.

    • NetworkPolicy
    • CNI
  3. 03

    Admission gated

    Admission control enforces what policy says at the only point it can be enforced completely: privileged pods, unsigned images and host mounts are rejected before they exist.

    • Admission policy
  4. 04

    Identity per workload

    Each workload gets its own identity to the cloud and to other services — short-lived, issued by the platform — so no long-lived key sits in a manifest waiting to be copied.

    • Service accounts
    • Workload identity

The audit log that leaves the cluster

API-server audit logging is enabled, shipped off the cluster and watched — because every reversal above is only provable, and every abuse of the remaining privilege only visible, if the record exists somewhere an attacker inside the cluster cannot edit.

Every stage runs in monitor mode before it enforces, so the hardening programme is never the cause of the outage it exists to prevent.

The responsibility line

What a managed control plane takes off your plate, and what it quietly leaves

Running Kubernetes through a cloud provider's managed service moves real work across the line: the control plane is patched, its components are operated, and etcd is protected without your team touching any of it. What the marketing leaves quieter is everything that stays on your side: who holds which RBAC role, whether the pod network is segmented, what admission allows, how workloads authenticate to the cloud, whether audit logging is even enabled — on some platforms it is not, until someone turns it on. A managed cluster with default settings is a well-patched control plane in front of a permissive estate.

The practical use of the responsibility line is that it tells you where to spend. There is no point auditing the provider's side beyond reading their attestations; there is every point in auditing yours, because that is where the findings live. In the clusters we assess, the recurring weaknesses sit entirely on the customer side of the line — access nobody narrowed, traffic nobody declared, logs nobody shipped — and none of them require a different provider to fix. They require decisions.

The recurring findings

Cluster habits that turn one compromised pod into an estate

None of these is exotic. Each is a convenience adopted early that nobody went back for, and each hands an intruder the next step for free.

  1. 01

    The CI credential that owns the cluster

    The deployment pipeline was granted cluster-admin on day one because deploys kept failing with less. Years later, whoever compromises the pipeline — or reads its stored credential — administers everything. A deploy needs the verbs of a deployer, in the namespaces it deploys to.

  2. 02

    The pod network with no walls

    With no network policies, the payments service is reachable from the marketing site's pod, and a compromise anywhere can probe everywhere. Segmentation in Kubernetes is not bought; it is declared, namespace by namespace, and until it is declared it does not exist.

  3. 03

    The secret that was only encoded

    A Kubernetes Secret is base64-encoded, which is a format, not a protection: anyone who can read the object reads the credential. Encryption at rest, tight RBAC on secret objects and platform-issued workload identity are what actually stand between an intruder and the keys.

  4. 04

    The API server nobody is watching

    Every act of administration and every abuse of it goes through the API server, and a cluster whose audit log is disabled — or kept only on the cluster itself — cannot answer the first question of any investigation: who did what, when. Enabling, shipping and watching that log is cheap before the incident and impossible after it.

Our role

What we change in the cluster and what your platform team keeps running

  1. 01

    The assessment before the change

    We map effective access — who can really do what, through every role binding and group — and observe real traffic before proposing a single denial, because a hardening plan built on assumptions becomes an outage plan.

  2. 02

    Enforcement that arrives gradually

    Policies run in observe mode first, exceptions are negotiated with the owning teams, and each enforcement step keeps a rollback — so trust in the programme grows with every step that breaks nothing.

  3. 03

    A cluster your team can defend

    The role model, the policy set and the reasons behind each are documented and handed over, with your engineers making the last changes themselves — because a hardened cluster that only outsiders understand softens on its own.

Which teams get which reach, and how much deploy friction the business accepts, are your decisions; we sequence the reversals so each one lands without stopping the delivery the cluster exists to carry.

Common questions

What buyers ask

Our Kubernetes cluster is managed by the cloud provider — how much security work is actually left for us?+

Most of it, measured by where incidents actually start. The provider patches and operates the control plane, which removes a real class of risk. Everything that decides whether a compromise spreads stays with you: RBAC and who effectively holds admin, network segmentation between workloads, what admission control allows into the cluster, how workloads authenticate to cloud services, secrets handling, node and image hygiene, and whether audit logs are enabled and shipped. In practice, the findings in managed-cluster assessments sit almost entirely on the customer side of that line — which is good news, because it means they can be fixed without migrating anywhere.

Where should Kubernetes RBAC hardening start when everyone currently has cluster-admin?+

With a map, not a decree. First establish who effectively holds what, through every binding, group and service account — the answer usually surprises the team that granted it. Then build the small set of roles that describe real jobs: deployer in these namespaces, reader across the estate, operator of this platform component. Move people and pipelines onto those roles while the old bindings still exist, watch for what breaks, then remove standing admin last, leaving a break-glass path that is logged and alerts when used. Revoking first and mapping later feels faster and costs more, because every failure becomes an emergency re-grant that outlives the emergency.

Are Kubernetes network policies worth the effort in a small cluster?+

Yes, and the small cluster is the cheap time to do it. The effort scales with the number of traffic relationships, so a cluster with a dozen services needs a map a person can draw in an afternoon; the same estate two years later needs a project. The security case does not shrink with size — a flat pod network makes every workload a bridge to every other, and the workloads that matter most are usually in the same cluster as the ones exposed most. Start with default-deny in the namespaces that hold anything sensitive, declare the flows those services actually use after observing them, and let new services arrive with their policies rather than retrofitting forever.

Next step

Assess your cluster with us

A cluster assessment that lists who effectively holds cluster-admin today, which traffic paths are open that nobody declared, and the hardening order that fixes both without an outage.