SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has introduced its Open Agent Safety Platform aimed at bolstering security measures for autonomous AI agents. This comprehensive system addresses agent testing, deployment, software, computing infrastructure, and robotics. On September 28, NVIDIA revealed the technology as companies increase their deployment of agents within complex computing environments. It integrates open source runtime software with hardware monitoring independent of the software environment, allowing organizations to tailor component deployment based on their security and operational needs.
The platform’s core component, OpenShell, functions as a secure runtime environment that establishes enforceable boundaries around AI agents. This open source software tracks agent activities and enforces policies while agents operate on CPUs. Administrators can specify access permissions related to files, networks, tools, processes, and credentials before an agent begins its tasks. During operation, OpenShell continuously enforces these defined limits. NVIDIA states that the software supports both open and closed AI models, providing a security boundary external to the model and its associated agent harness.
The second key element, NVIDIA Sentry, offers autonomous monitoring via BlueField-4 data processing units. Sentry functions outside the agent’s software environment, constantly observing agent behavior. If an agent attempts to breach its authorized limits, Sentry can quarantine it within milliseconds. This hardware-based system independently enforces security policies. Built on NVIDIA’s DOCA software, Sentry includes features such as agent identity verification, access controls, telemetry, and inspection of agent requests and responses.
Runtime security measures go beyond the AI model itself
By connecting application software, secure runtime controls, and computing infrastructure, the platform forms a multi-layered security architecture. OpenShell isolates agents within sandboxed environments with kernel-level separation and policy enforcement. Sentry adds an additional hardware security layer that agents cannot control directly. NVIDIA highlighted recent security breaches where agents bypassed application-level controls during task execution. The system is designed to keep controls separate from the AI models and their guiding algorithms, offering organizations enhanced security and control.
More than 100 organizations are utilizing technologies linked to the Open Agent Safety Platform, according to NVIDIA. Anthropic has collaborated with NVIDIA on implementing advanced controls for managed AI agents and sandbox environments. Additionally, Salesforce has integrated OpenShell with Slack to enable team members to monitor agent activities, audit events, and permissions. The platform’s users span sectors like financial services, enterprise software, cybersecurity, cloud infrastructure, robotics, and energy. It also offers security controls for autonomous systems functioning within physical environments.
Open source software broadens deployment possibilities
OpenShell operates alongside NVIDIA Vera, the CPU tailored for agent-driven AI workloads. NVIDIA provides options for developers to extend the open source software to third-party platforms, including systems utilizing Arm and Intel architectures. The entire platform can run across compatible hardware environments, while Sentry uses BlueField-4 for its autonomous monitoring functions. NVIDIA confirmed that its Open Agent Safety Platform software—including OpenShell and related capabilities—is accessible via its developer resources and the GitHub platform.
Jensen Huang, NVIDIA’s founder and CEO, emphasized that AI safety and security are comprehensive full-stack engineering challenges. The platform also contributes technology to the Open Secure AI Alliance, an initiative led by the Linux Foundation focused on open research, tools, and security practices for AI agents. NVIDIA stated that its platform supports governance across software, hardware, computing, and robotics layers. The launch introduces new runtime and hardware controls to assist organizations in deploying autonomous AI agents across an expanding array of business systems.