SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has introduced the Open Agent Safety Platform, a security system for autonomous artificial intelligence agents. The platform combines software controls with independent hardware monitoring across AI workloads. It covers development, testing and deployment, including systems used in enterprise computing and robotics. NVIDIA said the platform can enforce restrictions outside the AI model itself. Organizations can also select individual components based on their computing and security requirements.
OpenShell provides the platform’s open source runtime layer for controlling how AI agents interact with digital resources. Administrators can set rules for access to files, networks, tools, processes and credentials. The software records agent activity and enforces those permissions while tasks are running. OpenShell creates a security boundary around agents without relying on instructions inside the model. NVIDIA designed the software to work with both open and proprietary AI models.
NVIDIA Sentry adds a separate monitoring layer through the company’s BlueField-4 data processing units. The technology operates outside an agent’s main software environment and watches activity continuously. Sentry can isolate an agent within milliseconds when its actions cross established security limits. It can also enforce access policies without depending on the agent for compliance. NVIDIA uses its DOCA software framework to support identity checks, telemetry, inspection and other security functions.
Independent controls strengthen agent security
The architecture separates AI decision-making from the systems that enforce security rules. OpenShell places agents inside controlled runtime environments with isolation and policy enforcement. Sentry then monitors those environments from dedicated hardware that the agent does not manage. NVIDIA developed the platform after security researchers documented cases involving agents that moved beyond intended application controls. The company said external enforcement can give operators direct control over what agents can access and execute.
NVIDIA said more than 100 organizations are involved with technologies connected to the Open Agent Safety Platform. Participants include Anthropic, Microsoft, Cisco, CrowdStrike, Dell Technologies, HPE and Hugging Face. The group also includes JPMorganChase, Palantir, Palo Alto Networks, Salesforce, SAP and ServiceNow. Their work spans cybersecurity, cloud infrastructure, enterprise software and financial services. The platform also supports autonomous systems that interact with physical environments and require controls beyond application software.
Open source design broadens hardware support
OpenShell runs with NVIDIA Vera CPUs, which the company developed for demanding agentic AI computing workloads. Developers can also extend the open source runtime to systems based on third-party architectures. NVIDIA specifically identified Arm and Intel platforms among the supported expansion paths. Sentry relies on BlueField-4 hardware for its independent monitoring role. This arrangement separates runtime controls from the hardware layer responsible for observing and restricting agent activity across compatible computing environments.
NVIDIA founder and CEO Jensen Huang described AI safety and security as full-stack engineering challenges when announcing the platform. The company positioned Open Agent Safety Platform as a reference system for securing agents across software, hardware and infrastructure. OpenShell provides the software boundary, while Sentry supplies independent hardware oversight. Together, the components give operators tools to define permissions, record activity and enforce restrictions as autonomous AI agents move from development into operational computing environments.