SIAgentNews.com
← All news

NVIDIA opens source tools to secure and scale AI agents

NVIDIA is contributing new software to the Cloud Native Computing Foundation to help enterprises manage the security and infrastructure needs of autonomous AI agents.

· First reported by varindia

31 secondsReduced workload readiness time for GLM-5.3

The short version

  • NVIDIA is contributing OpenShell, a runtime for securing autonomous agents, to the Cloud Native Computing Foundation (CNCF).
  • The company is also supporting Agent Substrate, a Google-led framework for running long-running agent workloads on Kubernetes.
  • New tools like NVIDIA Snapshot can reduce AI model startup times from several minutes down to just a few seconds.
  • Hundreds of organizations have already adopted OpenShell, which has seen tens of thousands of contributions in under a year.

Why it matters

As companies move from simple chatbots to autonomous agents that can perform tasks, they need standardized ways to secure those agents. These tools provide a shared foundation for developers to build reliable, scalable, and secure AI systems without starting from scratch.

Opening the agent ecosystem

NVIDIA is shifting its AI agent infrastructure toward an open-source model by contributing several key projects to the Cloud Native Computing Foundation (CNCF). This move is designed to help the broader tech community manage the unique challenges of autonomous agents. By making these tools public, NVIDIA aims to create a shared foundation for how organizations deploy and scale AI across different environments.

The primary contribution is NVIDIA OpenShell, which serves as a runtime for governing and securing autonomous agents. OpenShell allows companies to apply deterministic controls, which are set rules that dictate what an agent can do. This ensures that as agents use tools or access private data, they remain within the boundaries set by the organization's security policies.

The company is also working on Agent Substrate, a framework led by Google. This project focuses on running long-running agent workloads on Kubernetes, which is a popular system for managing software containers. Together, these contributions provide a roadmap for how the next generation of cloud-native applications will handle complex, multi-step AI tasks.

Securing autonomous activity

One of the biggest hurdles for enterprise AI is ensuring that autonomous agents do not perform unauthorized actions. OpenShell addresses this by providing a layer of governance that integrates with existing compute resources and monitoring systems. It allows security professionals to define specific permissions for how agents interact with inference services and execute various workloads.

The project has already seen significant adoption, with hundreds of organizations using it since its launch. It has also gathered tens of thousands of contributions from the community in less than a year. This rapid growth suggests that the need for standardized agent security is a high priority for developers who are currently building these systems.

Jim Zemlin, CEO of the Linux Foundation, noted that the cloud-native community faces a new challenge in providing secure environments for agents. He stated that NVIDIA's contribution gives developers a shared foundation to shape how agent security evolves. This collaborative approach means that security standards will be developed by a wide range of organizations rather than a single provider.

Managing workloads at scale

Running AI agents at scale requires efficient management of hardware resources, particularly GPUs. NVIDIA is working with Agent Substrate to ensure compatibility with its own tools, such as NVIDIA Dynamo for model inference and KAI Scheduler for resource allocation. This connection helps link agent execution directly with the hardware that powers it.

A key feature of Agent Substrate is its ability to preserve agent environments between different tasks. Instead of keeping a dedicated computer running for every single agent session, the system can release compute resources when an agent is idle. When the agent needs to work again, the system can suspend and resume the environment, allowing for a much more efficient use of expensive hardware.

NVIDIA and Google are also collaborating on a specific integration between OpenShell and Agent Substrate. This partnership aims to provide a unified way to handle both security and governance across different environments. This includes both cloud-based systems and on-premises hardware, giving companies more flexibility in where they choose to host their AI operations.

Faster inference and deployment

To make AI agents feel responsive, they need to start up quickly. NVIDIA introduced several new components to its DSX AI Factory Platform to address this. These include NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR), which provide managed services for GPU infrastructure. These tools automate the lifecycle of a cluster and provide signed images to ensure software consistency.

Another specific tool, NVIDIA Snapshot, targets the delays caused by loading large models into memory. It works by restoring checkpointed replicas of models without having to repeat the entire loading or warmup process. This is particularly useful for scaling inference workloads or recovering from system interruptions, as it avoids the

What to watch

  • How other cloud providers will integrate NVKE and NVKR into their own services.
  • Whether OpenShell becomes the industry standard for agent governance.

Try it

Check the CNCF Sandbox page to see the status of the OpenShell and Agent Substrate projects.

Sources and links

First reported by varindia. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news