Google Agent Gateway Governs AI Agents on EKS, AKS

Google Cloud launched Agent Gateway for cross-cloud in Limited Preview on November 11, 2026, extending its agent governance infrastructure beyond Google Cloud Platform. The service brings centralized identity management and safety enforcement to AI agents running on CNCF-conformant Kubernetes clusters hosted on Amazon EKS, Azure AKS, Red Hat OpenShift, and bare-metal infrastructure.

What Agent Gateway Does

Agent Gateway establishes a unified control plane for AI agent operations across heterogeneous infrastructure. Each agent receives a verifiable identity registered in Agent Registry, Google's central catalog for autonomous systems. Local gateway proxies deployed within each cluster route three traffic types: LLM inference calls, Model Context Protocol (MCP) server requests, and agent-to-agent (A2A) communication. Those proxies enforce access policies and Model Armor safety rules before traffic leaves the cluster, then forward telemetry to Cloud Logging and Cloud Trace for centralized observability.

The architecture separates the data plane from the control plane. Agents execute within customer-managed Kubernetes clusters on AWS, Azure, or on-premises hardware, while Google's hosted services handle identity verification, policy distribution, and audit logging. Google states this design keeps sensitive workloads and model interactions within the customer's chosen cloud boundary while maintaining governance visibility.

Cross-Project Agent Discovery

A companion update shipped November 10, 2026 introduced cross-project agent discovery for organizations using Google Cloud's Resource Manager hierarchy. Accounts with the AGENT_MANAGEMENT capability can now enumerate agents deployed across member projects while preserving their runtime identities. This feature addresses multi-team deployments where individual business units provision their own agent fleets but central IT maintains security policy. The discovery mechanism surfaces agent metadata, deployment topology, and active policy bindings without requiring direct project access.

Release Date and Pricing

Officially launched on November 11, 2026, Agent Gateway for cross-cloud enters Limited Preview with no service fee. Google's release documentation notes that integrated products, including Cloud Logging, Cloud Trace, and any underlying compute or storage resources, continue to incur standard charges according to existing pricing schedules. Access requires submission of a formal request form, standard practice for Google Cloud preview programs with capacity constraints.

Competitive Context

The release arrives during a high-volume week for agent platform announcements tracked by industry feeds. Google's move to support Amazon EKS and Azure AKS reflects broader competition among hyperscalers to control the governance layer for multi-cloud AI deployments. By offering identity and safety services that work across rival clouds, Google positions Agent Gateway as infrastructure-agnostic middleware rather than a GCP-exclusive capability. The inclusion of bare-metal and OpenShift clusters signals intent to reach regulated industries and on-premises enterprise workloads where public cloud adoption remains partial.

What This Means

Agent Gateway for cross-cloud extends Google's agent governance stack into territories where it does not control the compute substrate. Organizations running AI agents on AWS or Azure gain access to centralized policy enforcement and audit logging without migrating workloads to Google Cloud. The Limited Preview serves as a test of demand for vendor-neutral agent governance tools. If adoption proves strong, expect Google to bundle Model Armor capabilities and Agent Registry into a broader multi-cloud AI management suite competing directly with platform-native offerings from AWS and Microsoft.