Breaking

NVIDIA Launches Open Agent Safety Platform to Govern AI Agents From Test to Deployment

September 28, 2026 at 14:38 EDT

AI Agents
Security
Infra & Chips

NVIDIA on Monday introduced NVIDIA Open Agent Safety Platform, an open software platform and reference system design meant to keep autonomous AI agents inside enforceable runtime boundaries as they move from testing into production. The system combines NVIDIA OpenShell, an open source runtime layer, with NVIDIA Sentry, a hardware-assisted monitoring design that runs on BlueField-4 DPUs and can quarantine agents in milliseconds when they attempt to move beyond allowed limits.

September 28, 2026 · NVIDIA

Agent safety moves outside the model

NVIDIA Open Agent Safety Platform combines OpenShell runtime controls with Sentry monitoring on BlueField-4 DPUs to keep AI agents contained, observable and auditable.

named ecosystem participants

80%

faster sandbox performance claimed for Vera

quarantine response target

Vera sandbox claim, visualized

1.0x

traditional CPU baseline

1.8x

Vera with OpenShell

NVIDIA says Vera delivers up to 80% faster sandbox performance, making broader agent containment less costly for production systems.

OpenShell

Sets runtime boundaries, traces actions and enforces policy around agent execution.

Sentry

Uses BlueField-4 DPUs for out-of-band monitoring and millisecond quarantine.

Control loop

Set policy

Define what agents may access.

→

Run agent

Execute inside a governed boundary.

→

Monitor

Inspect requests, identity and behavior.

→

Quarantine

Stop agents that cross boundaries.

Why teams like it

Runtime controls are auditable and enforceable outside prompts, useful for coding agents and enterprise workflows.

What remains open

The strongest isolation depends on BlueField hardware, and containment still needs model evaluation and human oversight.

Continue reading

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