# Google donates Agent Substrate and gVisor to the CNCF for securing AI agents

published: 2026-11-20T12:05:20.794Z
topic: infrastructure
source: https://www.infoq.com/news/2026/11/google-substrate-gvisor-cncf/?utm_campaign=infoq_content&amp;utm_source=infoq&amp;utm_medium=feed&amp;utm_term=AI%2C+ML+%26+Data+Engineering
url: https://aiagentnews.io/stories/google-donates-agent-substrate-and-gvisor-to-the-cncf-for-securing-ai-agents

Google is donating its Agent Substrate and gVisor projects to the Cloud Native Computing Foundation, putting two components of its agent infrastructure under vendor-neutral open-source governance. The move targets secure execution of AI-generated code and the compute overhead of agents that often pause and resume.

Google has announced it is donating Agent Substrate and gVisor to the Cloud Native Computing Foundation. The announcement also includes OpenRL, Google's open-source reinforcement learning orchestration project, which covers coordinating large-scale training workloads.

Google says agents can spend 90-95% of their time idle while still holding state such as memory, files, repositories and active sessions. Agent Substrate separates agent state, held in actors, from the compute used to run it, provided by reusable pre-warmed workers drawn from shared Kubernetes pools. According to Google, agents can resume in under 500 milliseconds without a new Kubernetes Pod for each execution.

On security, gVisor adds an isolation boundary through a user-space application kernel that intercepts system calls instead of letting containers reach the host Linux kernel directly. Agent Substrate also supports Cloud Hypervisor microVMs for workloads that need hardware-level virtualisation.

## Why it matters if you build agents

Teams running agents that execute generated code can look at gVisor or Cloud Hypervisor microVMs as isolation options without changing the agent lifecycle model.

Agents that sit idle between tool calls and model inference may be suspended and restored on pooled workers rather than holding a dedicated Pod, which reduces control plane load during large evaluation runs.

The projects are now under CNCF governance rather than Google alone, though the source does not describe a release timeline or adoption plan.
