Author says Kubernetes NetworkPolicy fails to isolate multiplexed AI agents

The author argues that Kubernetes NetworkPolicy is inadequate for isolating multiplexed AI agents because it operates at the pod or worker pool level rather than the individual actor level. To achieve true zero-trust isolation, security architectures must combine worker-level backstop policies with actor-identity enforcement at the egress proxy, tool-call guardrails, and request-level identity minted at restore and dropped at suspend. Open article →