NVIDIA Open Agent Safety Platform: How It Keeps AI Agents Secure
AI agents are becoming more capable of doing tasks on their own. They can use tools, access files, interact with applications and continue working with less human intervention. But as their capabilities increase, controlling what an AI agent is allowed to do becomes equally important.
To address this challenge, NVIDIA announced its Open Agent Safety Platform on September 28, 2026. The platform combines software and hardware-based controls to create additional security boundaries around AI agents.
Why Do AI Agents Need Security?
A normal chatbot mainly responds to user prompts. An AI agent can go further by using tools, accessing resources and performing actions.
For example, an AI coding agent might need access to a project folder, development tools and certain network services. Giving it unrestricted access could create unnecessary security risks.
This is where agent security becomes important. Instead of expecting the AI agent to control itself, security controls can be placed outside the agent itself.
NVIDIA's approach is designed around this idea.
What Is NVIDIA Open Agent Safety Platform?
NVIDIA's platform brings together two major components:
NVIDIA OpenShell — a secure runtime for running AI agents in isolated environments.
NVIDIA Sentry — an additional monitoring and enforcement layer designed to run on NVIDIA BlueField-4 DPUs.
Together, they create a layered approach for controlling AI-agent activity across software and infrastructure.
What Does OpenShell Do?
OpenShell is an open-source secure runtime that runs agents inside isolated sandboxes.
It can define what an agent is allowed to access, including files, networks, tools, processes and credentials. NVIDIA says OpenShell also monitors system calls and applies policies while the agent is running.
For developers, this means an AI agent does not automatically receive unlimited access to the entire system.
For example, if an agent only needs access to a particular project folder, its environment can be restricted instead of giving it access to unrelated files.
What Is NVIDIA Sentry?
Sentry adds another layer of protection.
According to NVIDIA, Sentry is an out-of-band watchdog that runs on BlueField-4 DPUs. It continuously monitors agent behavior and can enforce security policies independently of the agent's software environment.
If an agent attempts to move outside its defined boundaries, NVIDIA says Sentry can quarantine it within milliseconds.
This creates a separation between the AI agent and the system responsible for monitoring it.
Why Is This Important for Developers?
As businesses start using AI agents for coding, customer support, automation and other workflows, agents may receive access to increasingly important systems and data.
Security therefore cannot depend only on the AI model following instructions correctly.
NVIDIA's platform takes a zero-trust style approach, where permissions and boundaries are defined around the agent and monitored during execution.
This could be particularly relevant for developers building applications that allow AI agents to interact with databases, APIs, files or other business systems.
The Bigger Picture
The launch shows how AI-agent development is expanding beyond just improving model intelligence.
As agents become more autonomous, developers also need to think about permissions, isolation, monitoring and enforcement.
NVIDIA's Open Agent Safety Platform is one approach to this problem, combining an open-source runtime with an additional hardware-based monitoring layer.
For companies building AI-powered software, the message is straightforward: giving an AI agent more capabilities also means putting clear boundaries around what it can access and do.
As agentic AI continues to grow, security may become an important part of the infrastructure behind these systems—not just a feature added after development.
Need help with your digital project? Get in touch with Tech Assistant — we build solutions that scale.
