SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has launched the Open Agent Safety Platform, a security framework designed for autonomous artificial intelligence agents. It integrates software controls with hardware-based monitoring to oversee AI workloads across various stages including development, testing, and deployment—covering systems used in enterprise computing and robotics. NVIDIA explained that the platform can impose restrictions beyond the AI model itself, allowing organizations to choose specific components aligned with their security and computing needs.
The platform’s open source runtime layer, OpenShell, manages how AI agents access digital resources by establishing rules for files, networks, tools, processes, and credentials. It logs agent activities and enforces permissions during task execution. OpenShell creates a security perimeter around agents independently of instructions embedded within the models. NVIDIA designed the software to be compatible with both open-source and proprietary AI models.
Adding a separate oversight layer, NVIDIA Sentry operates through the company’s BlueField-4 data processing units. The technology functions outside the primary software environment of the agent, continuously monitoring activity. Sentry can isolate an agent within milliseconds if its behavior exceeds predefined security limits. It also enforces access policies independently of the agent’s compliance. NVIDIA’s DOCA software framework supports security functions such as identity verification, telemetry, and inspection.
Separate controls enhance agent security
The design separates AI decision-making from the enforcement of security policies. OpenShell ensures agents operate within controlled runtime environments that provide isolation and policy enforcement. Meanwhile, Sentry monitors these environments from dedicated hardware that agents do not manage. The platform was developed in response to security research highlighting cases where agents exceeded intended application controls. NVIDIA emphasized that external enforcement mechanisms offer operators direct oversight over what agents can access and perform.
NVIDIA reports that over 100 organizations are engaged with technologies related to the Open Agent Safety Platform. Notable participants include Anthropic, Microsoft, Cisco, CrowdStrike, Dell Technologies, HPE, and Hugging Face. The group also involves JPMorganChase, Palantir, Palo Alto Networks, Salesforce, SAP, and ServiceNow, covering fields such as cybersecurity, cloud infrastructure, enterprise software, and financial services. The platform also supports autonomous systems that interact with physical environments and require additional controls beyond software alone.
Open source approach broadens hardware compatibility
OpenShell operates on NVIDIA Vera CPUs, specifically developed for intensive agentic AI processing tasks. Developers can extend the open source runtime to other systems based on third-party architectures, with support explicitly identified for Arm and Intel platforms. For its monitoring functions, Sentry relies on the BlueField-4 hardware, which separates runtime control from the hardware layer responsible for observing and restricting agent activities across compatible computing environments.
Jensen Huang, NVIDIA’s founder and CEO, characterized AI safety and security as comprehensive full-stack engineering challenges during the platform’s announcement. The company positioned the Open Agent Safety Platform as a foundational reference system for securing agents across software, hardware, and infrastructure layers. OpenShell offers the software boundary, while Sentry provides independent hardware oversight. Collectively, these components enable operators to establish permissions, log activities, and enforce restrictions as autonomous AI agents transition from development to operational environments.