SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has unveiled its Open Agent Safety Platform aimed at enhancing security protocols for autonomous AI agents. This new system encompasses agent testing, deployment, software, computing frameworks, and robotics technology. NVIDIA revealed the technology on September 28, as enterprises increasingly integrate agents into complex computing environments. The platform combines open source runtime software with hardware monitoring that operates independently. NVIDIA stated that organizations can tailor the deployment of individual components to meet their security and operational needs.
OpenShell functions as the platform’s secure runtime layer, establishing enforceable boundaries around AI agents. This open source software monitors agent activity and enforces policies while agents operate on CPUs. Administrators are able to specify access controls for files, networks, tools, processes, and credentials before an agent begins its tasks. During operation, OpenShell ensures these limits are maintained. NVIDIA explained that the software supports both open and closed AI models, offering a security perimeter outside the model and its associated agent harness.
The second key component, NVIDIA Sentry, offers autonomous monitoring via BlueField-4 data processing units. Sentry functions outside the agent’s software environment, consistently observing agent activity. When an agent attempts to exceed its authorized boundaries, the system can quarantine it within milliseconds. Hardware-based security controls are also enforced independently. Built on NVIDIA’s DOCA software, Sentry enables agent identity verification, access management, telemetry, and inspection of request and response data.
Runtime controls extend beyond the model
This platform integrates application software, secure runtime mechanisms, and computing infrastructure into a multi-layered security framework. OpenShell places agents within sandboxed environments, utilizing kernel-level isolation and policy enforcement. Sentry adds a separate hardware layer that agents cannot manipulate directly. NVIDIA highlighted that recent security breaches demonstrated agents bypassing application-level controls during task execution. The system was designed to give organizations control measures that remain separate from both the AI agent and the controlling model.
More than 100 organizations are currently working with technologies linked to the Open Agent Safety Platform, according to NVIDIA. Anthropic has partnered with the company to develop additional safeguards for managed AI agents and sandboxed environments. Salesforce has integrated OpenShell with Slack, enabling teams to monitor agent activity, audit events, and permission requests. The platform’s users also span sectors such as financial services, enterprise software, cybersecurity, cloud infrastructure, robotics, and energy. Additionally, the platform supports security measures for autonomous systems operating within physical environments.
Open source tools facilitate broader deployment options
OpenShell is compatible with NVIDIA Vera, the company’s CPU optimized for agentic AI tasks. NVIDIA also allows developers to extend the open source software to third-party platforms, including systems powered by Arm and Intel technologies. The entire platform can run across compatible hardware environments, with Sentry utilizing BlueField-4 for its independent monitoring functions. NVIDIA announced that the Open Agent Safety Platform software, including OpenShell and related functionalities, is accessible through its developer resources and the GitHub platform.
NVIDIA founder and CEO Jensen Huang emphasized that the company regards AI safety and security as comprehensive engineering challenges. The platform also contributes to the Open Secure AI Alliance, an initiative overseen by the Linux Foundation. This alliance concentrates on open research, security tools, and best practices for AI agents. NVIDIA stated that its platform supports governance across software, hardware, computing, and robotics layers. The new release introduces additional runtime and hardware controls to assist organizations deploying autonomous AI agents across an expanding range of business applications.