← Blog
#Kubernetes#DevOps#Security#Cloud Infrastructure#AI Workloads

Kubernetes v1.36: Security Defaults Tighten as AI Workload Support Matures

webhani·

Kubernetes v1.36 landed in May 2026 with 70 enhancements. Three features graduated to General Availability — User Namespaces, Mutating Admission Policies, and Fine-Grained Kubelet API Authorization — and Ingress NGINX was officially retired back in March. Here's what matters for teams running production clusters.

User Namespaces: Stronger Container Isolation

User Namespaces reached GA in v1.36. The feature allows container processes to run under UIDs/GIDs that are mapped to unprivileged host users, even if they appear as root inside the container.

Without User Namespaces, a container escape that grants root access inside the container could grant the same effective access on the host. With it, the blast radius shrinks considerably.

# Enable User Namespaces at the Pod level
apiVersion: v1
kind: Pod
metadata:
  name: secure-workload
spec:
  hostUsers: false  # Enable User Namespaces
  containers:
  - name: app
    image: my-app:latest
    securityContext:
      runAsUser: 0  # Root inside the container, but mapped to
                    # an unprivileged UID on the host

For security-sensitive production workloads, this should be part of your baseline Pod spec. Check that your container images don't rely on host-level UID assumptions before enabling it.

Mutating Admission Policies: Webhooks Without the Overhead

Mutating Admission Policies are now GA, enabling CEL-based resource mutation without a dedicated Webhook server. For common mutation use cases — adding default labels, injecting sidecar configs — this reduces both operational overhead and a potential failure point.

apiVersion: admissionregistration.k8s.io/v1alpha1
kind: MutatingAdmissionPolicy
metadata:
  name: inject-team-label
spec:
  matchConstraints:
    resourceRules:
    - apiGroups: [""]
      apiVersions: ["v1"]
      operations: ["CREATE"]
      resources: ["pods"]
  mutations:
  - patchType: "ApplyConfiguration"
    applyConfiguration:
      expression: |
        Object{
          metadata: Object.metadata{
            labels: {"team": "platform"}
          }
        }

This doesn't replace all Webhook use cases — complex logic requiring external state or multi-step decisions still needs a Webhook. But for declarative mutations, the native approach is cleaner.

Fine-Grained Kubelet API Authorization

Kubelet API access is now more granular. Previously, access was relatively coarse — you either had it or you didn't. The new authorization model allows scoping access to specific subresources, reducing the attack surface for lateral movement within a cluster.

Review your existing RBAC policies for anything that grants broad Kubelet API access and tighten accordingly.

The Ingress NGINX Retirement

Ingress NGINX was retired on March 24, 2026. This is the most impactful change for many clusters.

Your migration options are:

  1. Gateway API — the Kubernetes-native successor, now stable and recommended for new deployments
  2. Envoy Gateway — a Gateway API implementation backed by Envoy, CNCF-graduated
  3. NGINX Ingress Controller from NGINX Inc. — a separate (non-community) NGINX-based implementation
# Check which Ingress controller version you're running
kubectl get pods -n ingress-nginx \
  -o jsonpath='{.items[*].spec.containers[*].image}'
 
# Check if Gateway API CRDs are already installed
kubectl get crd gateways.gateway.networking.k8s.io 2>/dev/null \
  && echo "Gateway API present" || echo "Not installed"

The retirement means no new features and eventually no security patches. Start migration planning now rather than waiting for an incident to force the decision.

AI Workloads and What's Changing

Kubernetes is increasingly the default orchestration substrate for GPU-based workloads. The traditional "golden signals" — CPU and memory — need to expand to include GPU utilization, memory bandwidth, and inference throughput.

Platform Engineering teams are finding that abstractions like Ray or Kubeflow are necessary to prevent ML engineers from needing deep Kubernetes expertise just to run training jobs.

Summary

v1.36 is a meaningful release for security posture. The GA graduation of User Namespaces and Fine-Grained Kubelet Auth gives teams concrete handles to tighten their baseline. The Ingress NGINX retirement is the immediate action item — audit your clusters and start evaluating Gateway API if you haven't already.

← Back to Blog