Breaking

NVIDIA Launches Open Agent Safety Platform to Govern Autonomous AI Agents

September 28, 2026 at 08:59 EDT

AI Agents
Security
Infra & Chips

NVIDIA has launched NVIDIA Open Agent Safety Platform, an open software platform and reference system design meant to secure autonomous AI agents from testing through deployment. The system combines NVIDIA OpenShell, open source software for policy-governed agent runtimes, with NVIDIA Sentry, a reference design that uses NVIDIA BlueField-4 DPUs to monitor agents from outside their own execution environment.

Sept. 28, 2026 · NVIDIA

Agent safety moves into the stack

NVIDIA Open Agent Safety Platform pairs OpenShell runtime controls with Sentry monitoring on BlueField-4 DPUs to govern autonomous agents from testing through deployment.

100+

organizations joining the platform effort

core layers: OpenShell and Sentry

quarantine response described in milliseconds

Partner scale, drawn proportionally

1single organization

100+platform partners

Each orange block represents roughly ten organizations, showing why the launch is being positioned as an ecosystem move rather than a single-product update.

How the safety loop works

Define policy

Set access to files, tools, credentials and networks.

→

Run in OpenShell

Trace actions and enforce runtime boundaries.

→

Monitor with Sentry

Observe from BlueField-4 and quarantine unsafe behavior.

Enterprise upside

Supporters value identity-bound agents, least-privilege access, audit trails, runtime policy checks and faster containment.

Open questions

Teams are watching real deployment depth, hardware dependence and portability across mixed Arm, Intel and cloud environments.

Continue reading

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