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 hardware monitoring with software controls across various AI tasks, including development, testing, and deployment phases. It applies to enterprise computing and robotics, allowing restrictions to be enforced beyond the AI model itself. Organizations have the flexibility to pick specific components aligned with their security and computing needs.
OpenShell serves as the platform’s open-source runtime layer, managing how AI agents access digital assets. Administrators are empowered to establish rules governing access to files, networks, tools, processes, and credentials. During operation, OpenShell logs agent activities and enforces permissions, creating a security perimeter around agents independent of internal model instructions. The software is compatible with both open and proprietary AI models, as NVIDIA intended.
In addition, NVIDIA Sentry provides an additional layer of security via the company’s BlueField-4 data processing units. The technology operates outside of an agent’s core software environment, maintaining continuous activity surveillance. Sentry can isolate an agent within milliseconds if its behavior exceeds predefined security parameters. Moreover, it can enforce access policies without relying on the agent’s compliance. NVIDIA’s DOCA software framework supports functions like identity verification, telemetry, and security inspection within Sentry.
Separate controls reinforce agent protection
The system architecture decouples AI decision-making from the security enforcement mechanisms. OpenShell embeds agents within controlled runtime environments, ensuring isolation and policy adherence. Meanwhile, Sentry monitors these environments via dedicated hardware that the agents do not control. NVIDIA developed this platform after security experts reported incidents where agents bypassed application controls. The external enforcement mechanism provides operators with direct oversight over agents’ access and activities, enhancing security.
Over 100 organizations are reported to be engaged with technologies related to the Open Agent Safety Platform. 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 collective efforts span cybersecurity, cloud infrastructure, enterprise applications, and financial services. The platform additionally caters to autonomous systems operating in physical environments that demand more than software controls.
Open source architecture promotes wider hardware compatibility
OpenShell functions on NVIDIA Vera CPUs, engineered specifically for complex agentic AI workloads. The open source runtime can also be adapted for third-party architectures, with NVIDIA explicitly citing Arm and Intel platforms as supported options. Sentry’s independent monitoring role depends on BlueField-4 hardware, separating runtime controls from the hardware layer responsible for observing and restricting agent behavior across compatible systems. This structure ensures a clear division between control and observation layers.
Jensen Huang, NVIDIA’s founder and CEO, characterized AI safety and security as full-stack engineering challenges when unveiling the platform. The Open Agent Safety Platform is positioned as a reference design for securing agents across software, hardware, and infrastructure components. OpenShell establishes the software boundary, while Sentry delivers hardware-level oversight. Together, these elements give operators the tools to set permissions, track activities, and enforce restrictions, facilitating safe deployment of autonomous AI agents from development through operational deployment.