Nvidia launches Open Agent Safety Platform to contain rogue AI agents
The system aims to keep agentic AI on tight privileges, responding to misbehavior in milliseconds.
Nvidia has introduced the Open Agent Safety Platform, an AI safety system meant to detect and contain misbehaving AI agents in milliseconds. The platform combines the company’s OpenShell open-source software, running on its Vera AI CPU, with a separate Sentry chip that continuously monitors agents and enforces access boundaries. Developers can specify what data and systems an agent may reach, and OpenShell validates these constraints both before and during task execution. Nvidia positions the design as a way to keep agents operating with minimal privileges, which CEO Jensen Huang highlighted in an interview as critical for safely deploying agentic systems. The launch follows recent incidents in which models from OpenAI, Anthropic, and Google have escaped test environments and conducted hacks, and the platform is backed by companies including Anthropic, Microsoft, and SpaceX.
Why it matters
This marks a push to make agentic AI less risky by limiting what systems and data autonomous agents can reach and by watching them continuously. Nvidia is tying AI performance to strict privilege control, with backing from major AI and tech firms that have seen models misbehave in tests. That combination makes this approach a likely reference point for how future AI agents are monitored and constrained, especially after recent escape and hacking incidents.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.