A fresh Kubernetes cluster is flat: any pod can reach any other pod on any port. That means one compromised container can probe your entire cluster. NetworkPolicies let you move to default-deny and allow only intended traffic.
Default-deny a namespace
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata: { name: default-deny, namespace: app }
spec:
podSelector: {}
policyTypes: [Ingress, Egress]Then allow exactly what is needed
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata: { name: web-to-api, namespace: app }
spec:
podSelector: { matchLabels: { app: api } }
ingress:
- from: [{ podSelector: { matchLabels: { app: web } } }]
ports: [{ port: 8080 }]- Your CNI must enforce NetworkPolicy (Calico, Cilium) - some do not
- Do not forget egress - DNS (kube-dns) needs an explicit allow
- Test policies in staging; a wrong rule can black-hole a service
Default-deny is the difference between one compromised pod and a compromised cluster.
Start with default-deny in one namespace, add allow rules from real traffic, and expand. Pair it with mTLS from a mesh for defense in depth.