SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has launched the Open Agent Safety Platform, a comprehensive security framework designed for autonomous artificial intelligence agents. This system integrates software controls with independent hardware oversight across various AI workloads, spanning development, testing, and deployment phases, including enterprise computing and robotics. NVIDIA explained that the platform is capable of imposing restrictions beyond the AI model itself, allowing organizations to choose specific components tailored to their computing and security needs.
OpenShell serves as the platform’s open source runtime layer, managing how AI agents access digital resources. Administrators can establish rules governing access to files, networks, tools, processes, and credentials. The software tracks agent activities and enforces permissions during task execution. OpenShell creates a security perimeter around agents without relying on embedded instructions within the models. It is designed to function seamlessly with both open and proprietary AI models, according to NVIDIA.
NVIDIA Sentry provides an additional monitoring layer, utilizing the company’s BlueField-4 data processing units. The technology operates independently from an agent’s core software environment, continuously observing activity. When an agent’s actions breach predefined security parameters, Sentry can isolate it within milliseconds. Furthermore, it enforces access policies without relying on the agent’s compliance. NVIDIA employs its DOCA software framework to facilitate identity verification, telemetry, inspection, and various other security functions.
Independent security controls enhance protection for AI agents
The system architecture separates AI decision-making processes from security enforcement mechanisms. OpenShell embeds agents within controlled runtime environments, ensuring isolation and policy enforcement, while Sentry monitors these environments through dedicated hardware that remains unmanaged by the agents. NVIDIA developed the platform in response to security research documenting incidents where agents exceeded their intended application controls. The company emphasizes that external enforcement allows operators to have direct oversight of agent access and execution capabilities.
More than 100 organizations are reportedly engaged with technologies related to the Open Agent Safety Platform. Key participants include Anthropic, Microsoft, Cisco, CrowdStrike, Dell Technologies, HPE, and Hugging Face. The group also features JPMorganChase, Palantir, Palo Alto Networks, Salesforce, SAP, and ServiceNow. Their collaborative efforts span cybersecurity, cloud infrastructure, enterprise software, and financial services sectors. Additionally, the platform supports autonomous systems interacting with physical environments that require controls beyond software alone.
Open source approach enhances compatibility with various hardware platforms
OpenShell operates on NVIDIA Vera CPUs, engineered specifically for demanding agentic AI computing tasks. Developers also have the option to extend the open source runtime to systems utilizing third-party architectures. NVIDIA identified Arm and Intel platforms as supported options for expansion. Sentry depends on BlueField-4 hardware to perform its independent monitoring function, effectively segregating runtime controls from the hardware layer responsible for observing and limiting agent activity across compatible computing environments.
Jensen Huang, NVIDIA’s founder and CEO, described AI safety and security as full-stack engineering challenges during the platform’s announcement. The company positioned the Open Agent Safety Platform as a reference architecture for securing agents across software, hardware, and infrastructure layers. OpenShell establishes the software boundary, while Sentry provides hardware-based oversight. These components collectively equip operators with tools to set permissions, monitor activities, and enforce restrictions as autonomous AI agents transition from development to operational deployment in computing environments.