SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has launched the Open Agent Safety Platform, a security solution designed for autonomous artificial intelligence agents. The system integrates software controls with independent hardware monitoring across various AI workloads, encompassing development, testing, and deployment phases. It is applicable to enterprise computing and robotics systems. NVIDIA explained that the platform is capable of imposing restrictions beyond the AI model itself. Organizations can choose specific components aligned with their security and computing needs.
OpenShell functions as the platform’s open source runtime layer, managing how AI agents access digital resources. Administrators are able to establish rules governing access to files, networks, tools, processes, and credentials. The software logs agent activities and enforces permissions throughout task execution. OpenShell creates a security boundary around agents without relying on internal instructions within the models. NVIDIA designed this software to be compatible with both open-source and proprietary AI models.
NVIDIA Sentry offers an additional monitoring layer, utilizing the company’s BlueField-4 data processing units. The technology functions outside the main software environment of an agent, continuously observing activity. Sentry can isolate an agent within milliseconds if its actions breach predefined security parameters. It can also enforce access policies independently of the agent’s compliance. NVIDIA employs its DOCA software framework to facilitate identity checks, telemetry, inspection, and other security functionalities.
Independent controls bolster security for AI agents
The system architecture separates AI decision-making processes from the enforcement of security policies. OpenShell confines agents within controlled runtime environments, ensuring isolation and policy enforcement. Meanwhile, Sentry continuously monitors these environments using dedicated hardware that the agents do not manage. The platform was developed in response to security researchers documenting incidents where agents exceeded their intended application controls. NVIDIA noted that external enforcement mechanisms provide operators with direct oversight of what agents can access and perform.
More than 100 organizations are reported to be involved with technologies related to the Open Agent Safety Platform. Participants include Anthropic, Microsoft, Cisco, CrowdStrike, Dell Technologies, HPE, and Hugging Face. The collaboration also involves JPMorganChase, Palantir, Palo Alto Networks, Salesforce, SAP, and ServiceNow. Their efforts span cybersecurity, cloud infrastructure, enterprise software, and financial services. The platform also supports autonomous systems that interact with physical environments, requiring controls beyond application software.
Open source architecture expands hardware compatibility
OpenShell operates with NVIDIA Vera CPUs, designed specifically for demanding agentic AI computing tasks. Developers have the ability to extend the open source runtime to third-party systems, with support identified for Arm and Intel architectures among potential options. Sentry relies on BlueField-4 hardware to perform its independent monitoring functions. This setup separates runtime control from the hardware layer responsible for observing and limiting agent activity across compatible computing environments.
NVIDIA founder and CEO Jensen Huang emphasized that AI safety and security are comprehensive full-stack engineering challenges when introducing the platform. The company presented the Open Agent Safety Platform as a reference model for securing agents across software, hardware, and infrastructure layers. OpenShell delivers the software boundary, while Sentry provides hardware-based independent oversight. Together, these components enable operators to define permissions, log activities, and enforce restrictions as autonomous AI agents transition from development to operational environments.