SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has rolled out its Open Agent Safety Platform to bolster security measures for autonomous AI agents. The new system encompasses agent testing, deployment, software, computing infrastructure, and robotics. NVIDIA revealed the technology on September 28 as businesses increase their deployment of agents within complex digital environments. It integrates open source runtime software with standalone hardware monitoring solutions. According to NVIDIA, organizations have the flexibility to implement individual components of the platform based on their specific security and operational needs.
OpenShell functions as the secure runtime environment for the platform, establishing enforceable boundaries around AI agents. This open source software tracks agent activity and enforces policies while agents operate on CPUs. Operators can specify permissions related to files, networks, tools, processes, and credentials prior to an agent’s execution. During operation, OpenShell ensures these restrictions are maintained. NVIDIA claims that the software supports both open and closed AI models and creates a security boundary external to the model and its agent harness.
The second core component, NVIDIA Sentry, offers independent monitoring via BlueField-4 data processing units. Sentry functions outside the agent’s software environment, continuously observing its behavior. If an agent attempts to exceed its authorized boundaries, Sentry can quarantine it within milliseconds. The system enforces security policies independently through hardware. Built on NVIDIA’s DOCA software, Sentry supports features such as agent identity verification, access controls, telemetry, and inspection of agent requests and responses.
Runtime governance extends beyond the AI model itself
The new platform integrates application software, secure runtime controls, and computing infrastructure within a layered security architecture. OpenShell isolates agents inside sandboxed environments with kernel-level separation and policy enforcement. Additionally, Sentry introduces a separate hardware layer that agents cannot directly manipulate. NVIDIA highlighted recent security breaches where agents bypassed application controls during task execution. The system was designed to provide organizations with controls that are distinct from the AI agent and the guiding model itself.
Over 100 organizations are currently working with technologies connected to the Open Agent Safety Platform, NVIDIA reports. Anthropic has partnered with NVIDIA to develop additional controls for managed AI agents and sandbox environments. Salesforce has integrated OpenShell with Slack, enabling teams to review agent activity, audit logs, and permission requests. Participants also include sectors such as financial services, enterprise software, cybersecurity, cloud infrastructure, robotics, and energy. The platform further supports security controls for autonomous systems operating in physical environments.
Open source solutions broaden deployment opportunities
OpenShell is compatible with NVIDIA Vera, the company’s CPU designed specifically for agent-driven AI workloads. NVIDIA also permits developers to expand the open source software to third-party computing platforms, including those utilizing Arm and Intel architectures. The entire platform can operate across compatible hardware environments, with Sentry relying on BlueField-4 for its independent monitoring functions. NVIDIA states that the software for the Open Agent Safety Platform, including OpenShell and related tools, is accessible via its developer resources and the GitHub platform.
NVIDIA founder and CEO Jensen Huang emphasized that the company considers AI safety and security as comprehensive full-stack engineering challenges. The platform also contributes to the Open Secure AI Alliance, an initiative overseen by the Linux Foundation that promotes open research, tools, and security standards for AI agents. NVIDIA added that its platform supports governance across software, hardware, computing, and robotics layers. The release introduces new runtime and hardware controls as organizations expand their deployment of autonomous AI agents across an increasing array of business systems.