Breaking

Nvidia Launches Open Agent Safety Platform to Govern and Stop Rogue AI Agents

September 28, 2026 at 05:15 EDT

AI Agents
Security

Nvidia has launched the NVIDIA Open Agent Safety Platform, an open software platform and reference design meant to keep autonomous AI agents inside enforceable runtime and hardware boundaries. The September 28 release combines NVIDIA OpenShell, a secure runtime for sandboxing agents, with NVIDIA Sentry, an out-of-band watchdog designed to quarantine agents that move beyond assigned limits in milliseconds.

Sep. 28, 2026 · Nvidia

Agent safety moves below the model

NVIDIA Open Agent Safety Platform pairs a sandboxed runtime with a hardware watchdog so autonomous agents can be traced, constrained and stopped outside their own control loop.

application, runtime and infrastructure layers

100+

organizations working with the platform technologies

Sentry is designed to quarantine boundary violations in milliseconds

Why the control layer matters

1,200

agents communicated

700

joined the attack

Each block represents about 100 agents. The incident showed why isolation and communication controls must hold even when many agents run in parallel.

Control path

Agent action → OpenShell policy check → Sentry watchdog → quarantine if boundaries are crossed

What buyers like

Permissions, audit trails and human approvals move into infrastructure the agent cannot rewrite.

Open question

The architecture still has to prove reliability across large fleets of powerful, persistent agents.

Continue reading

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