Introducing Agent Compute Engine and expanded agent governance
Google Cloud introduces Agent Compute Engine and expanded agent governance to securely scale autonomous AI agents across environments, addressing infrastructure and security challenges for enterprise deployment.
Google Cloud is launching Agent Compute Engine, a new compute service designed to securely run autonomous AI agents at scale. Built to handle high-churn, iterative workloads like code execution and browser automation, it provides hardware-isolated runtime sandboxes for each agent. The service will enter preview in the coming months, offering a secure foundation for running custom agents alongside existing workloads. The infrastructure aims to address the unique demands of agentic workflows that traditional compute models were not optimized to support.
To enable multi-cloud portability and customization, Google is introducing Agent Substrate on Google Kubernetes Engine (GKE), an open-source execution runtime. It decouples agent execution from machine management using a zero-trust architecture, snapshotting idle agents to preserve resources. Leading AI teams like Nous Research are already leveraging this architecture to deploy multi-tenant agents efficiently within customer infrastructure, ensuring per-agent isolation and extensible access control.
Addressing security risks in growing agent fleets, Google is expanding its Gemini for business agent governance capabilities across Google Cloud runtimes, cross-cloud, on-premises, and client devices. The update introduces new controls to manage agent sprawl, including shadow AI and data exfiltration risks, by enforcing least-privilege access and auditing asynchronous actions. Governance features will now span diverse environments to maintain security and compliance as agent deployments scale.
New agent governance capabilities are also being added to Agent Gateway, including granular cost controls, intelligent model routing, and simplified customizations. These updates aim to provide unified identity, policy, and observability across runtimes such as Cloud Run. Companies like L’Oréal are adopting these tools to scale agentic AI while maintaining control over agent access and actions, ensuring secure innovation across their platforms.