Most programs labeled Zero Trust are network segmentation projects with a new name. Real Zero Trust is an identity-centric reorganization of authorization: every request, from every principal, to every resource, is evaluated against current context — device posture, location, behavior, risk score — and granted the minimum capability required.
First moves that compound
Three changes consistently produce outsized results: short-lived credentials for service-to-service authentication, device posture as a first-class policy input, and per-request authorization for sensitive APIs. None require a platform replatform; all require political will to retire long-lived secrets and standing access.
Agents as principals
An AI agent inside your VPC with a long-lived service account is functionally a privileged insider. Zero Trust extends naturally to autonomous principals: per-agent identity, capability tokens scoped to a single tool, valid for minutes, and continuously attested by the orchestrator. Segmentation slows lateral movement; identity changes the blast radius itself.