NVIDIA Launches Open Agent Safety Platform — Here’s How It Protects AI Agents
NVIDIA has launched a new open platform designed to give organizations more control over AI agents as they move from testing environments into real-world deployment.
Called the NVIDIA Open Agent Safety Platform, the system combines open-source software with a reference system design intended to enforce security boundaries across the software, computing hardware and robotics systems that AI agents use to perform tasks.
The platform brings together NVIDIA OpenShell and NVIDIA Sentry, with NVIDIA positioning the combination as a way to provide full-stack governance and control for increasingly autonomous AI systems.
The announcement comes as companies are giving AI agents responsibility for longer-running and more complex tasks, increasing the need for security controls that exist beyond the AI model itself.
What Is the NVIDIA Open Agent Safety Platform?
NVIDIA Open Agent Safety Platform is designed to provide security controls across multiple layers of an agentic AI system rather than relying exclusively on the model or application layer.
The platform includes OpenShell, an open-source secure runtime that establishes boundaries around how AI agents execute tasks on CPUs.
According to NVIDIA, OpenShell can trace agent actions and enforce policies while an agent is running. The company says the software provides an enforceable security boundary outside the model and the agent harness.
OpenShell is now broadly available and is designed to run with NVIDIA Vera, NVIDIA’s purpose-built CPU for agentic AI. Because OpenShell is open source, NVIDIA says it can also be extended to work with third-party compute platforms, including those based on Arm and Intel technologies.
The second major component is NVIDIA Sentry, a reference system design built around NVIDIA BlueField-4 DPUs.
Sentry operates as an out-of-band watchdog that continuously monitors agent activity. NVIDIA says that if an agent attempts to move outside its permitted software boundary, Sentry can quarantine and stop the agent in milliseconds.
How NVIDIA Sentry Adds Another Security Layer
A central feature of the platform is that Sentry operates independently from the agent itself.
NVIDIA says Sentry uses an isolated, out-of-band trust domain to monitor activity, enforce policies and protect data. The approach is intended to prevent an AI agent from bypassing the controls responsible for supervising it.
The system combines threat detection, hardware-based governance and enforcement, identity verification, telemetry and granular access controls.
Sentry is built on NVIDIA DOCA, the company’s software framework for programmable BlueField DPU capabilities.
According to NVIDIA, DOCA allows Sentry to inspect agent requests and responses, provide attested telemetry, verify agent identity and enforce zero-trust policies governing access to data, tools, APIs and services.
That architecture is intended to create security controls that remain outside the agent’s direct control.
Why NVIDIA Is Focusing on Agent Security
AI agents are increasingly being designed to do more than generate text or answer questions. They can interact with software, access tools, retrieve information and carry out multi-step tasks.
That creates a different security challenge from traditional AI applications.
NVIDIA says recent security incidents have highlighted a recurring pattern in which an agent circumvents application-layer controls while attempting to complete an assigned task.
The company’s approach is therefore based on adding security boundaries at multiple levels of the technology stack.
“AI’s extraordinary potential for society will only be realized if we solve AI safety,” NVIDIA founder and CEO Jensen Huang said.
Huang said the company believes safety and security require full-stack engineering and that the new platform is intended to bring industry, researchers and public-sector organizations together around best practices, evaluation methods and international cooperation.
Anthropic, SpaceXAI and Other Companies Join NVIDIA
NVIDIA is also building the platform around a broader industry ecosystem.
The company says organizations across AI infrastructure, cybersecurity, enterprise software, robotics and financial services are working with its agent safety technologies.
Anthropic and NVIDIA have collaborated on additional security controls for agent deployments. Anthropic’s Claude Managed Agents establish a security boundary by separating the agent loop from the sandboxes where its work executes. NVIDIA says integrations with OpenShell and BlueField can provide additional controls over agent access to those environments.
Anthropic Chief Commercial Officer Paul Smith said companies need to be able to direct and verify what AI agents do, particularly when agents are operating in sensitive environments.
SpaceXAI is using NVIDIA Open Agent Safety Platform for Cursor coding agents and Grok models. The company said customers should be able to establish limits for agents and trust that those limits will remain enforced outside the model.
Scale AI is working with NVIDIA to incorporate the platform’s technologies into the agentic infrastructure layer of its Scale GenAI Portfolio.
Other organizations named by NVIDIA include Salesforce, SAP, Microsoft, Palantir, Perplexity, CrowdStrike, Cisco, Hugging Face, ServiceNow, Palo Alto Networks, Red Hat, IBM and JPMorganChase, among many others.
NVIDIA said more than 100 organizations are working with its Open Agent Safety Platform technologies.
The company also said robotics organizations including Figure, Gecko Robotics and Skild AI are building with OpenShell, extending agent safety controls to autonomous systems that can take action in physical environments.
OpenShell Expands Into Enterprise Workflows
NVIDIA’s announcement also points to integrations designed to make agent oversight part of existing enterprise workflows.
Salesforce and NVIDIA have integrated OpenShell with Slack, allowing teams to view agent activity and audit events and approve or reject requests for additional permissions.
SAP is embedding OpenShell into its Joule Studio runtime as part of the SAP Business AI Platform. SAP is also contributing engineering work to OpenShell and working with NVIDIA on interoperability standards through the Open Secure AI Alliance.
The broader ecosystem includes infrastructure providers and technology companies offering systems that support NVIDIA Open Agent Safety Platform technologies.
These include Dell Technologies, HPE, Lenovo, Oracle Cloud Infrastructure, Supermicro, CoreWeave, Microsoft, Cisco, Baseten, Together AI and others, according to NVIDIA.
NVIDIA Open Agent Safety Platform Availability
NVIDIA says its Open Agent Safety Platform software, including OpenShell and related skills, is available through NVIDIA developer resources and GitHub.
The company also said the platform contributes to the broader mission of the Open Secure AI Alliance, an initiative started by NVIDIA alongside more than 120 organizations and governed by the Linux Foundation.
The alliance is focused on open research, tools and shared approaches to AI agent security, including projects such as the Shared AI Findings Exchange, or SAFE.
For NVIDIA, the launch represents an attempt to make AI-agent security a full-stack infrastructure problem rather than something handled only inside an AI model or application.
As organizations deploy agents with greater access to software, data and physical systems, the question will increasingly be how those systems can remain useful while operating within clearly enforced boundaries.
NVIDIA’s new platform is designed to provide one answer: put controls around the agent at the runtime, hardware and infrastructure levels, and make those controls harder for the agent itself to bypass.
