NVIDIA Launches Open Agent Safety Platform to Secure Agents From Testing to Deployment
NVIDIA announced the NVIDIA Open Agent Safety Platform, an open software platform and reference system design that combines NVIDIA OpenShell secure runtime software and the NVIDIA Sentry reference system to provide full-stack governance and control for AI agents across software, compute and robotics systems. The company says OpenShell is broadly available and Sentry runs on BlueField-4 DPUs to monitor and quarantine agent behavior in milliseconds; NVIDIA cites collaboration with over 100 organizations.

What NVIDIA announced
NVIDIA announced the NVIDIA Open Agent Safety Platform as an open software platform and reference system design intended to strengthen AI security from agent testing to deployment. The company frames the platform as enabling full-stack governance and control across the software that runs agents, the hardware and compute layers, and robotics systems that execute tasks in the physical world. NVIDIA positions the platform as modular, allowing organizations to deploy elements according to their requirements.
Sources: S1
Platform components and technical scope
The platform comprises two principal elements described by NVIDIA: OpenShell, a secure runtime that sets enforceable boundaries for agents running on CPUs, and Sentry, an out-of-band watchdog implemented as a reference system design that runs on BlueField-4 DPUs. NVIDIA states OpenShell is broadly available, runs with "minimal overhead" on the Vera CPU, and can be extended to third-party compute platforms. Sentry is described as an in-silicon enforcement layer that continuously monitors agent activity and can quarantine agents that attempt to move outside their software boundary in milliseconds.
Sources: S1
Ecosystem participation
NVIDIA named a range of industry participants and contributors across AI, cloud, security, finance, energy and robotics. The release cites collaborations and integrations with specific vendors and services, and states that "over 100 organizations" are working with the platform technologies. NVIDIA also links the platform to the Open Secure AI Alliance, which it says was initiated alongside "over 120 leading organizations" and is governed by the Linux Foundation.
Sources: S1
Availability and positioning
NVIDIA states that Open Agent Safety Platform software, including OpenShell and skills, are available through NVIDIA developer resources and GitHub. NVIDIA frames the initiative as supporting broader AI safety and security work through open research, skills and tools and cites integrations with a range of infrastructure and software partners.
Sources: S1
Why this may matter (inference)
If the platform operates as described, it offers organizations an additional set of tools to impose runtime boundaries, monitor agent activity, and enforce policies across hardware and software layers. For enterprises deploying long-running or agentic systems, the availability of an open-source runtime plus a hardware-level watchdog could provide more auditable controls and integration points with existing infrastructure and governance processes.
Sources: S1
What remains uncertain
- All technical claims, performance characteristics and partner commitments are reported by NVIDIA in a company press release and are not independently verified in this draft.
- The release does not provide quantitative benchmarks or independent test results for OpenShell overhead on NVIDIA Vera or for Sentry quarantine latency beyond the descriptive phrase "in milliseconds."
- The nature and depth of integration, production readiness, and contractual commitments from listed partners are not detailed in the release and vary by organization.
Sources
- NVIDIA Launches Open Agent Safety Platform to Secure Agents From Testing to DeploymentNVDA Press Releases · 2026-09-28