Technology

Google brings Agent Substrate to GKE for isolated agent sandboxes

The open-source runtime is designed to provision high-density sandboxes for AI agents, with isolation and suspend-resume capabilities.

An isolated AI agent sandbox running inside a secure GKE container runtimeAn isolated AI agent sandbox running inside a secure GKE container runtime

Google Cloud announced Agent Substrate availability on Google Kubernetes Engine on September 15. The open-source runtime is designed to run high-density agent sandboxes, with isolation and suspend-resume features intended for workloads that create many short-lived or stateful environments.

Agents that can execute code or interact with tools need boundaries around files, credentials, network access and compute. A substrate built on Kubernetes could make those environments easier to provision at scale, though the security of any deployment still depends on configuration and threat modeling.

Infrastructure for agentic software

The move reflects a shift from serving model requests to hosting entire agent sessions. Each session may need a workspace, dependencies and a safe place to run actions. Efficiently suspending and restoring that state could reduce resource waste while keeping the interaction responsive.

Isolation needs careful review

An open-source runtime is not a security guarantee on its own. Operators should inspect the project’s isolation model, patching practices, permissions and network controls, then test whether boundaries hold under realistic workloads before trusting it with sensitive tasks.

Sources

Google CloudGKEAI agents
FD

Written by

Fieldnote DeskTechnology brief at Fieldnote
About the editors