NVIDIA Introduces Open Platform for AI Agent Security
NVIDIA has introduced the NVIDIA Open Agent Safety Platform, an open software platform and reference system design aimed at providing security controls for AI agents from testing through deployment. The platform combines NVIDIA OpenShell, open-source runtime software designed to establish boundaries around agent activity, with NVIDIA Sentry, a reference system design that uses NVIDIA BlueField-4 data processing units (DPUs) to monitor agent behaviour and enforce policies independently of the agent itself.
NVIDIA said the platform is intended to provide controls across the software used to run AI agents, the computing infrastructure supporting them and, in some cases, robotics systems that allow agents to interact with the physical world. “As we continue to discover the frontier of AI capabilities, we must accelerate discovery at the frontier of AI safety,” said Jensen Huang, founder and CEO of NVIDIA. “Safety and security require full-stack engineering. NVIDIA Open Agent Safety Platform brings together industry, researchers and public-sector organisations to share best practices, align on evaluation methods and foster international cooperation.”
Runtime controls for AI agents
OpenShell provides a runtime environment in which policies can be applied to the actions performed by AI agents. NVIDIA said the software is designed to provide a security boundary outside the model and agent framework, allowing organisations to control how agents interact with systems and resources.
The software is available as open source and can be extended to work with third-party computing platforms, including Arm and Intel systems. NVIDIA said OpenShell is optimised to run on its Vera CPU, which was designed for agentic AI workloads.
The company said the platform is intended to address scenarios in which an AI agent attempts to bypass application-level security controls while carrying out an assigned task.
Hardware-based monitoring
The platform’s reference system design includes NVIDIA Sentry, an out-of-band monitoring mechanism that runs on NVIDIA BlueField-4 DPUs. Sentry is designed to monitor agent activity independently of the main computing environment and enforce policies at the hardware level. According to NVIDIA, it can isolate an agent if it attempts to operate outside its defined software boundaries.
The system uses NVIDIA DOCA software to inspect agent requests and responses, provide telemetry, verify agent identity and apply access policies covering data, tools, APIs and services.
Industry adoption
NVIDIA said more than 100 organisations are working with technologies associated with the Open Agent Safety Platform. Among them, Anthropic is integrating OpenShell and BlueField with its Claude Managed Agents architecture. Anthropic said the approach separates the agent loop from the environments in which tasks are executed.
“Companies are giving AI agents more of their most important work, and they need to direct and verify what those agents do, especially in sensitive environments,” said Paul Smith, chief commercial officer of Anthropic. Scale AI is also working with NVIDIA to incorporate the platform’s technologies into the infrastructure supporting its GenAI Portfolio. The company said it is using the reference design to provide isolation, policy enforcement and auditability for agentic AI applications.
Salesforce has integrated OpenShell with Slack, allowing teams to view agent activity and audit events and approve or reject requests for additional permissions. SAP is integrating OpenShell with the Joule Studio runtime within its SAP Business AI Platform and contributing engineering work to the OpenShell project.
Other organisations identified by NVIDIA as working with the platform or related technologies include Accenture, Armadin, Cadence, Cisco, Cognition, CrowdStrike, Dassault Systèmes, Deloitte, EY, Hugging Face, IBM, Microsoft, Palantir, Palo Alto Networks, Perplexity, ServiceNow, Siemens, Synopsys and others. NVIDIA also said robotics companies including Figure, Gecko Robotics and Skild AI are using OpenShell in work involving autonomous systems.
Financial services organisations including Citi and JPMorganChase are collaborating with NVIDIA on open-source agent security technologies, while energy companies including Hitachi Energy, EPRI, NextEra Energy, Quanta Services, Schneider Electric, Siemens Energy and Worley are among organisations working with the platform. Infrastructure software companies including Canonical, SUSE and Red Hat are also integrating OpenShell and related technologies into their platforms.
The NVIDIA Open Agent Safety Platform is positioned by the company as an open approach to applying security and governance controls across the AI agent stack, rather than relying solely on controls within individual AI models or applications.
