SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has launched the Open Agent Safety Platform, a comprehensive security solution designed specifically for autonomous artificial intelligence agents. This platform integrates software controls with dedicated hardware monitoring to oversee AI workloads. It is applicable throughout the entire lifecycle, covering development, testing, and deployment, including systems used in enterprise computing and robotics. NVIDIA emphasized that the platform can implement restrictions beyond the AI model itself, giving organizations the flexibility to choose components based on their specific computing and security needs.
OpenShell functions as the open-source runtime layer that manages how AI agents interact with digital resources. Administrators have the ability to establish rules governing access to files, networks, tools, processes, and credentials. The software keeps track of agent activities and enforces permissions during task execution. Notably, OpenShell establishes a security boundary around agents without relying on embedded instructions within the models. NVIDIA designed this software to be compatible with both open-source and proprietary AI models, ensuring broad applicability.
NVIDIA Sentry provides an additional layer of monitoring through the company’s BlueField-4 data processing units. The technology operates independently from the agent’s main software environment, continuously observing activity. When an agent’s actions exceed predefined security thresholds, Sentry can isolate it within milliseconds. Furthermore, it can enforce access policies autonomously, without relying on the agent for compliance. NVIDIA employs its DOCA software framework to facilitate identity verification, telemetry, inspection, and other security functionalities, ensuring robust protection.
Decoupled controls enhance agent security measures
This architecture distinctly separates AI decision-making from the enforcement of security protocols. OpenShell operates by placing agents inside controlled runtime environments that enforce policies and ensure isolation. Meanwhile, Sentry supervises these environments through dedicated hardware that the agent itself does not manage. NVIDIA developed this platform in response to documented cases by security researchers where agents moved beyond their intended operational controls. The company asserts that external enforcement mechanisms provide operators with direct authority over what agents can access and execute, significantly strengthening security.
NVIDIA reports that over 100 organizations are currently engaged with technologies related to the Open Agent Safety Platform. Participants include Anthropic, Microsoft, Cisco, CrowdStrike, Dell Technologies, HPE, and Hugging Face. The collaborative efforts also involve JPMorganChase, Palantir, Palo Alto Networks, Salesforce, SAP, and ServiceNow. These organizations work across sectors such as cybersecurity, cloud infrastructure, enterprise software, and financial services. The platform further extends support to autonomous systems that interact with physical environments, necessitating controls beyond traditional application software.
Open source architecture broadens hardware compatibility
OpenShell operates on NVIDIA Vera CPUs, which the company specifically developed to handle demanding agentic AI computing tasks. Additionally, developers have the ability to extend the open-source runtime to other systems based on third-party architectures. NVIDIA has identified Arm and Intel platforms as supported options for expansion. Sentry relies on BlueField-4 hardware for its independent monitoring function, effectively separating runtime control from the hardware layer tasked with observing and restricting agent activities across various compatible computing environments.
Jensen Huang, founder and CEO of NVIDIA, characterized AI safety and security as full-stack engineering challenges during the platform’s announcement. The company introduced the Open Agent Safety Platform as a reference architecture that ensures security across software, hardware, and infrastructure layers. OpenShell provides the software-enforced boundary, while Sentry offers independent hardware oversight. The integrated approach gives operators the tools to define permissions, log activities, and impose restrictions as autonomous AI agents transition from development to operational deployment in complex computing environments.