SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has launched the Open Agent Safety Platform, a robust security framework designed for autonomous artificial intelligence agents. This system integrates software controls with dedicated hardware monitoring for various AI workloads. It is applicable across stages of development, testing, and deployment, including applications in enterprise computing and robotics. NVIDIA emphasized that the platform can impose restrictions beyond the AI model itself, allowing organizations to customize component choices based on their specific security and computing needs.
OpenShell serves as the open-source runtime layer of the platform, managing how AI agents access digital resources. It enables administrators to establish rules governing file, network, tool, process, and credential access. During operation, the software logs agent activities and enforces permissions. OpenShell constructs a security boundary around agents independently of the instructions within the models, making it compatible with both open and proprietary AI systems, according to NVIDIA.
NVIDIA Sentry introduces an additional security layer through the company’s BlueField-4 data processing units. The technology operates outside the primary software environment of the agents, continuously monitoring their activities. When an agent’s actions breach security thresholds, Sentry can isolate it within milliseconds. It also enforces access policies without relying on the agent for compliance, leveraging NVIDIA’s DOCA software framework to support functions such as identity verification, telemetry, and inspection.
Decoupled controls bolster agent security
The platform architecture distinctly separates AI decision-making processes from the security enforcement systems. OpenShell places agents within controlled runtime environments that feature isolation and policy management. Meanwhile, Sentry oversees these environments using dedicated hardware, which the agents do not control. NVIDIA developed this platform following reports from security researchers about agents that exceeded their intended application controls. The company states that external enforcement mechanisms provide operators with direct authority over what agents can access and execute.
NVIDIA reports that over 100 organizations are engaging with technologies related to the Open Agent Safety Platform. Among the participants are Anthropic, Microsoft, Cisco, CrowdStrike, Dell Technologies, HPE, and Hugging Face. The collaboration also includes JPMorganChase, Palantir, Palo Alto Networks, Salesforce, SAP, and ServiceNow. These entities operate across sectors such as cybersecurity, cloud infrastructure, enterprise software, and financial services. The platform further supports autonomous systems interacting with physical environments that require security controls beyond software alone.
OpenShell’s open-source design enables compatibility with NVIDIA Vera CPUs, which are engineered for intensive agentic AI workloads. Developers can also extend its capabilities to third-party architectures, with NVIDIA specifically supporting Arm and Intel platforms for expansion. Sentry depends on BlueField-4 hardware for its independent monitoring function, effectively separating runtime controls from the hardware responsible for overseeing and restricting agent activities across different computing environments.
Jensen Huang, NVIDIA’s founder and CEO, highlighted AI safety and security as comprehensive full-stack engineering challenges at the platform’s announcement. The company introduced the Open Agent Safety Platform as a reference model to secure agents across hardware, software, and infrastructure layers. OpenShell provides the boundary for software control, while Sentry offers independent hardware-based oversight. This integrated approach equips operators with tools to manage permissions, record activities, and enforce restrictions as autonomous AI agents transition from development to operational environments.