Google Cloud has rolled out Agent Identity as a first-class principal type in IAM, built on the open SPIFFE standard rather than a proprietary scheme. The mechanics are specific: when you deploy an agent, Google Cloud assigns it a unique SPIFFE identity and an X.509 certificate, and each certificate is valid for 24 hours, with Google Cloud automatically keeping it current. That short expiry window matters more than it sounds. It replaces the old pattern of shared service accounts and long-lived keys — credentials that get copied, forgotten in config files, and reused across VMs indefinitely. A commercial write-up of the launch notes that a certificate rotates automatically every 24 hours and is cryptographically tied to the token, so a stolen token alone accomplishes nothing. The identity isn't just for authentication — it's built for attribution. All end-user access events are also attributable to the agent's SPIFFE ID, enabling easy governance, and the system integrates with audit logging to ensure accountability and provide clear audit logs both when the agent is acting as itself and when it is acting on behalf of an end user. That distinction — agent acting alone versus agent acting on a user's behalf — is one most agent identity schemes have struggled to represent cleanly, and Google handles it through a token-binding mechanism where a user's delegated JWT carries a cryptographic thumbprint tied to the agent's own certificate, so a leaked token can't be replayed by a different agent. Google also extended the identity into network policy: VPC Service Controls support for Agent Identity as first-class principals in ingress and egress rules is now in preview, letting administrators block data exfiltration attempts by binding perimeter rules directly to an agent's SPIFFE identity rather than to a shared account or IP range. The choice of SPIFFE is the interesting part. It's a CNCF-graduated standard already used for workload identity in Kubernetes environments, not something Google invented for this launch. That means the same trust-domain and namespace model can, in principle, federate across clouds — an agent identity minted on one platform could be recognized by policy on another, if other providers adopt the same substrate. Whether that federation actually happens, or whether every cloud vendor ships its own flavor of agent identity that only talks to itself, is the open question the next year of enterprise deployments will answer.