Breaking

NVIDIA Launches Open Agent Safety Platform to Set Runtime Boundaries for AI Agents

September 28, 2026 at 09:42 EDT

Security
AI Agents
Infra & Chips

NVIDIA launched NVIDIA Open Agent Safety Platform on September 28, 2026, framing it as an open software platform and reference system design for controlling what autonomous AI agents can access and do. The package combines OpenShell, an open source runtime boundary for agents, with Sentry, a BlueField-4 DPU-based watchdog designed to monitor behavior outside the agent's own environment and quarantine agents that try to move beyond permitted boundaries.

September 28, 2026 | NVIDIA

Safety Moves Below the Agent

NVIDIA Open Agent Safety Platform pairs an open runtime boundary with a silicon-based watchdog so operators can define, monitor and enforce what AI agents may access and do.

100+

organizations working with the platform

120+

Open Secure AI Alliance launch participants

80%

faster sandbox performance claimed for Vera

Ecosystem Scale

100+

Platform organizations

120+

Alliance organizations

Each block represents roughly 10 organizations, showing the partner effort beside the broader alliance.

Control Path

OpenShell

Sandbox and policy runtime

→

Policy Checks

Files, network, tools and credentials

→

Sentry

Out-of-band monitoring and quarantine

What backers value

Clear runtime boundaries, audit trails and enforcement outside the agent process.

What remains open

Actual deployment breadth and semantic judgment over multi-step agent behavior.

Continue reading

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