Breaking

NVIDIA Launches Open Agent Safety Platform to Contain AI Agents Across Software and Hardware

September 28, 2026 at 08:08 EDT

AI Agents
Security
Open Source

NVIDIA on September 28, 2026 introduced NVIDIA Open Agent Safety Platform, an open software platform and reference design meant to keep autonomous AI agents inside enforceable boundaries from testing through deployment. The system pairs the open-source NVIDIA OpenShell runtime with NVIDIA Sentry, a BlueField-4 DPU-based watchdog designed to monitor and stop agents from outside the software environment they can touch.

2026-09-28 · NVIDIA

Agent safety moves into the stack

NVIDIA's Open Agent Safety Platform combines an open runtime with a hardware watchdog so agent policy can be enforced outside the model and agent harness.

Platform collaborators

100+

organizations working with the platform technologies

Alliance scale

120+

organizations tied to the Open Secure AI Alliance

Security layers

application, runtime and infrastructure controls

Two open-security efforts at ecosystem scale

100+

platform organizations

120+

alliance organizations

The larger the column, the larger the collaboration footprint. Each block represents roughly ten organizations.

How the control path works

Application

models, tools, data and agent tasks

→

Runtime

OpenShell sandbox and policy checks

→

Infrastructure

Sentry on BlueField-4 monitors and can quarantine

Why it lands

Security moves outside the prompt. Policies, credentials, audit events and network paths are enforced by the runtime and, when used, a separate hardware trust domain.

What remains hard

The full watchdog layer depends on BlueField-4 systems, teams must integrate a new runtime, and telemetry settings require explicit operator decisions.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.