Breaking

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

September 28, 2026 at 18:26 EDT

AI Agents
Security
Infra & Chips

NVIDIA launched NVIDIA Open Agent Safety Platform on September 28, 2026, positioning it as an open software platform and reference system design for securing autonomous AI agents from testing through deployment. The package combines NVIDIA OpenShell, NVIDIA Sentry, NVIDIA BlueField-4, NVIDIA DOCA and NVIDIA Vera to move agent controls outside the model itself and into runtime and infrastructure layers.

Sep. 28, 2026 · NVIDIA

Safety Moves Outside the Agent

NVIDIA Open Agent Safety Platform combines OpenShell runtime controls with Sentry monitoring on BlueField-4 hardware to keep autonomous agents isolated, observable and governed.

100+

organizations working with platform technologies

120+

organizations tied to the Open Secure AI Alliance

80%

faster sandbox performance claimed for Vera

The Ecosystem Scale

Two adjacent communities show how widely NVIDIA is trying to distribute agent-safety controls.

100+

platform organizations

120+

alliance organizations

Control Loop

Sandbox

Isolate files, networks and processes.

→

Verify Policy

Check permissions before the agent runs.

→

Monitor Path

Inspect model calls and tool access.

→

Quarantine

Stop activity when boundaries break.

Why teams like it

Controls sit outside the model, can run on existing CPU infrastructure through OpenShell, and create audit trails for agent activity.

What remains hard

The strongest Sentry layer depends on BlueField-4, and verifiable policies still need precise engineering.

Continue reading

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