Breaking

NVIDIA Builds Open Agent Safety Platform With Industry Partners as Agent Security Concerns Rise

September 28, 2026 at 05:17 EDT

AI Agents
Security
Open Source

NVIDIA says it is building the NVIDIA Open Agent Safety Platform with partners across the industry, pitching the effort as a way to give AI agents stronger operating boundaries as enterprises hand them more consequential work. The move extends NVIDIA's 2026 agent-security push from open runtimes and shared incident reporting toward a broader platform layer intended to let companies deploy agents with more confidence.

Sep. 28, 2026 · NVIDIA

Agent Safety Moves to the Runtime

NVIDIA is building an Open Agent Safety Platform with industry partners as enterprises give AI agents more access to files, APIs, credentials and business systems.

120+

organizations in the open security alliance

17,000+

actions analyzed in a cited forensics case

>50%

claimed AI-Q query cost reduction

SAFE reporting clock

business days

day factual report

day remediation update

Each block represents about 10 days, with the first column compressed to one block for the initial confidential report window.

Boundary-first agent stack

Identity

Know which agent is acting.

→

Policy

Define what it may touch.

→

Runtime

Enforce limits outside the model.

Why teams like it

Hard runtime controls, signed skills and audit trails address risks that prompt-only guardrails cannot reliably contain.

What remains open

Voluntary incident reporting still faces legal and governance questions, especially for sensitive failures.

Continue reading

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