Advertise
AIComputers

Google Cloud Unveils Agent Compute Engine for AI Agents

Google Cloud's Agent Compute Engine runs AI agents in isolated microVMs with immutable storage and Titan chip firmware checks, rolling out in coming months.

Google Cloud announced Agent Compute Engine on October 23, 2026, an infrastructure service designed to run AI agents in isolated sandboxes. According to SiliconANGLE, each sandbox contains the applications, data and user instructions an agent needs, along with security features to block malicious activity.

The sandboxes are implemented as microVMs, which Google says use hardware as efficiently as containers while providing virtualization-level security. Google applied hypervisor technology that isolates virtual machines from one another so malware cannot spread between them.

The microVMs run on separately configured machines with a stateless design that does not permanently store files used by customer agents. Stored data sits in an immutable file system that cannot be modified or deleted, and Google verifies firmware integrity with its Titan security chip while applying side-channel attack mitigations to the CPU.

Google plans to deploy Agent Compute Engine along with a new version of its agent gateway within the next few months. That gateway update adds model routing, which sends each request to the model that can handle it most efficiently, and tools to control agent inference costs.

The new gateway also includes token management features that control token usage across agents and models for spend allocation and budget control.

Quick answers

What is Google Cloud's Agent Compute Engine?

It is an infrastructure service for running AI agents in isolated microVM sandboxes that contain the applications, data and user instructions the agent needs, plus security features to block malicious activity.

When will Agent Compute Engine be available?

Google plans to deploy it along with a new version of its agent gateway within the next few months.

How does Agent Compute Engine secure AI agents?

It uses microVMs with hypervisor isolation, a stateless design, an immutable file system, Titan security chip firmware verification and CPU side-channel attack mitigations.

Source