NVIDIA Launches Open Agent Safety Platform to Secure AI Agents
NVIDIA has launched the Open Agent Safety Platform, combining open-source software with a reference system design to provide security controls for AI agents across software, computing infrastructure and robotics. The platform was announced on September 28, 2026, and is designed to add enforceable boundaries as AI agents take on increasingly autonomous tasks.
The platform brings together NVIDIA OpenShell and NVIDIA Sentry, creating separate layers of control that can monitor and restrict agent activity. NVIDIA says more than 100 organisations are working with technologies from the platform, including Anthropic, Microsoft, Salesforce, SAP, Cisco, CrowdStrike and Palantir.
OpenShell Creates a Secure Runtime Boundary
OpenShell provides an open-source runtime environment that places AI agents inside sandboxed environments. It can enforce policies governing the files, networks, tools, processes and credentials an agent can access while carrying out a task.
The system is designed to keep these controls outside the AI model and agent harness, meaning an agent cannot simply modify its own security boundaries. OpenShell also logs policy decisions and can restrict network activity and API access according to predefined rules.
NVIDIA says OpenShell is now broadly available and can run on NVIDIA Vera CPUs. As open-source software, it can also be extended to work with third-party compute platforms based on Arm and Intel processors.
Sentry Adds Hardware-Level Monitoring
The second major component is NVIDIA Sentry, an out-of-band watchdog included in the platform's reference system design. Sentry runs on NVIDIA BlueField-4 DPUs, operating separately from the host system running the AI agent.
This separation allows Sentry to monitor agent activity from an independent trust domain. NVIDIA says that if an agent attempts to move beyond its permitted software boundary, Sentry can quarantine and stop it in milliseconds. The system uses NVIDIA DOCA to inspect agent requests and responses, verify identities, provide telemetry and enforce zero-trust policies.
Platform Targets Enterprise and Physical AI
NVIDIA is positioning the platform for AI agents operating across enterprise software, computing infrastructure and physical robotics. This becomes increasingly relevant as agents move beyond generating text or code and begin interacting with tools, APIs, databases and physical systems.
The company says the platform allows organisations to deploy individual components according to their security requirements rather than requiring a single configuration across every workload.
Major AI Companies Join the Ecosystem
NVIDIA says more than 100 organisations are working with the platform's technologies. Anthropic is collaborating with NVIDIA on integrating OpenShell and BlueField technologies with Claude Managed Agents, while Salesforce has integrated OpenShell with Slack to give teams visibility into agent activity and permission requests.
SAP is embedding OpenShell into its Joule Studio runtime, while companies including Cisco, CrowdStrike, Microsoft, Palantir, Palo Alto Networks, Scale AI and ServiceNow are also working with NVIDIA on agent-security technologies.
NVIDIA Pushes Security Beyond the AI Model
The launch reflects a shift toward securing AI agents at multiple layers rather than relying only on model-level safeguards. NVIDIA's approach places controls around the runtime and, through Sentry, adds an independent hardware enforcement layer.
OpenShell and related software are available through NVIDIA's developer resources and GitHub. The platform also contributes to the broader Open Secure AI Alliance, which NVIDIA initiated with more than 120 organisations and which is governed by the Linux Foundation.
