TLDR: Google has introduced two new open-source tools, Agent Sandbox and Multi-Tier Checkpointing (MTC), alongside GKE Pod Snapshots, a new feature for its managed Kubernetes service. These innovations aim to improve the security, efficiency, and performance of artificial intelligence environments, particularly for AI agent deployments and large-scale AI training projects. The announcement was made at the KubeCon + CloudNativeCon North America conference.
Atlanta, GA – November 11, 2025 – Google LLC today announced the debut of two significant open-source tools and a new feature for its Google Kubernetes Engine (GKE) service, all designed to streamline the operation and enhance the security of artificial intelligence environments. The unveiling took place at the KubeCon + CloudNativeCon North America conference in Atlanta.
The first new open-source tool, Agent Sandbox, addresses the critical cybersecurity risks associated with AI agents interacting with external applications. AI agents, which often utilize browsers and databases to perform tasks, can inadvertently or maliciously generate malware. To counter this, Agent Sandbox facilitates the creation of isolated container environments, or ‘sandboxes,’ that are cordoned off from sensitive systems. Implemented as an extension to Kubernetes’ core feature set, Agent Sandbox allows AI applications to launch thousands of isolated AI agent environments and automatically delete them upon task completion. This tool is built upon gVisor, an open-source technology released by Google in 2018, which isolates containers from the underlying operating system’s sensitive components, preventing malicious code from making unauthorized changes. Google Cloud will integrate and support Agent Sandbox within its GKE service, enabling developers to create ‘pre-warmed’ Agent Sandbox environments. These pre-initialized containers, equipped with necessary AI agent tools, come online swiftly, eliminating delays in agent workflows and speeding up processing.
Complementing Agent Sandbox, Google also introduced Multi-Tier Checkpointing (MTC), another open-source tool aimed at optimizing large-scale AI training projects. MTC is designed to streamline these complex operations, enhancing overall efficiency.
Alongside these open-source tools, Google Cloud debuted GKE Pod Snapshots, a new feature for its GKE managed Kubernetes service. This feature promises to significantly improve the performance of AI workloads, particularly for large language models (LLMs) that can take upwards of 10 minutes to launch. According to Brandon Royal, a senior product manager at Google, ‘GKE Pod Snapshots supports snapshot and restore of both CPU- and GPU-based workloads, bringing pod start times from minutes down to seconds.’ He further noted, ‘With Pod Snapshots, any idle sandbox can be snapshotted and suspended, saving significant compute cycles with little to no disruption for end-users.’ This capability allows for the suspension of idle sandboxes, conserving compute resources while ensuring rapid resumption of operations.
Also Read:
- Google Labs’ Jules: Autonomous AI Coding Agent Redefines Software Development
- Deutsche Telekom Launches AI-Powered RAN Guardian Agent for Autonomous Network Optimization
These new offerings underscore Google’s commitment to advancing AI infrastructure, focusing on both robust security measures and performance optimizations for developers working with AI agents and large-scale models. The announcement was made by Maria Deutscher of SiliconANGLE.
