SANTA CLARA, CALIFORNIA / RankWire.AI / – NVIDIA has introduced its Open Agent Safety Platform to bolster security measures for autonomous AI agents. This new system encompasses agent testing, deployment, software, computing infrastructure, and robotics systems. On September 28, NVIDIA revealed the technology as enterprises increase their deployment of agents within complex computing environments. The platform integrates open source runtime software with hardware monitoring capabilities independent of the software. NVIDIA stated that organizations can tailor the deployment of individual platform components to align with their security and operational needs.
OpenShell functions as the platform’s secure runtime environment, establishing enforceable boundaries around AI agents. As open source software, it monitors agent activity, enforces policies, and runs on CPUs. Administrators can set restrictions on access to files, networks, tools, processes, and credentials before an agent begins its tasks. During operation, OpenShell ensures these limitations are maintained. NVIDIA indicated that the software is compatible with both open and closed AI models and creates a security boundary outside the model and its agent harness.
The second key component, NVIDIA Sentry, offers independent oversight through BlueField-4 data processing units. Operating outside the agent’s software environment, Sentry continually observes agent behavior. It can quarantine an agent within milliseconds if it attempts to exceed its designated boundaries. Additionally, it enforces security policies through hardware. Sentry is built on NVIDIA’s DOCA software, which supports features such as agent identity verification, access controls, telemetry, and inspection of agent requests and responses.
Runtime controls extend beyond the model
The platform connects application software, secure runtime controls, and computing infrastructure into a layered security architecture. OpenShell places agents within sandboxed environments with kernel-level isolation and policy enforcement. Sentry adds a separate hardware layer inaccessible to agents directly. NVIDIA emphasized recent security incidents where agents bypassed application-level controls while executing tasks. The system was designed to ensure organizations retain controls separate from the AI agent and the model guiding its actions.
Over 100 organizations are currently working with technologies related to the Open Agent Safety Platform, NVIDIA reported. Anthropic has partnered with NVIDIA on additional security measures for managed AI agents and sandbox environments. Salesforce has integrated OpenShell with Slack, enabling teams to monitor agent activity, audit events, and permission requests. Participants span various sectors, including financial services, enterprise software, cybersecurity, cloud infrastructure, robotics, and energy. The platform also offers security features for autonomous systems operating in physical environments.
Open source software broadens deployment choices
OpenShell is compatible with NVIDIA Vera, the company’s CPU optimized for agentic AI workloads. NVIDIA also permits developers to extend the open source software to third-party computing platforms, such as those utilizing Arm and Intel architectures. The overall platform can function across compatible hardware environments, with Sentry utilizing BlueField-4 for its independent monitoring roles. NVIDIA stated that the Open Agent Safety Platform software, including OpenShell and associated tools, is accessible through its developer resources and the GitHub software development platform.
Jensen Huang, founder and CEO of NVIDIA, expressed that the company considers AI safety and security as comprehensive, full-stack engineering challenges. The platform also contributes to the Open Secure AI Alliance, an initiative led by the Linux Foundation. This alliance focuses on open research, tools, and security best practices for AI agents. NVIDIA affirmed that its platform supports governance across software, hardware, computing, and robotics layers. The release introduces additional runtime and hardware controls to assist organizations deploying autonomous AI agents across an expanding array of business systems.