The vulnerability lies in a misconfigured Kubernetes NetworkPolicy named "inter-ns" that was created by the application's deployment code. The advisory states that this policy, intended to provide network isolation for a 'hardened' namespace, was 'mis-written' and allowed pods within that namespace to communicate with pods in any other namespace, enabling potential lateral movement. The provided commit patch confirms this by removing the code responsible for creating the inter-ns NetworkPolicy.
The analysis of the patch 3bb5e9d9ce1199dfbb90fef8ad79ebdeb0bc5e78 shows the removal of a netwv1.NewNetworkPolicy resource from the provision method within the Hardening struct. This change is present in two separate but identical files: environment/deploy/parts/hardening.go and install/deploy/parts/hardening.go.
The removed code defined an egress policy that selected all pods in the namespace (PodSelector: {}) and allowed traffic (Egress rule) to any namespace whose name was not the current one (Operator: "NotIn"). While the developer's comment indicated an intent to 'deny all traffic to adjacent namespaces', the implementation did the opposite: it explicitly allowed egress traffic to all other namespaces. In a Kubernetes environment where network policies are used to enforce a default-deny stance, this rule effectively created a hole in the firewall.
The vulnerable functions are identified as parts.Hardening.provision in both locations, as they are directly responsible for executing the Pulumi code that defines and creates this flawed NetworkPolicy resource during deployment.