NVIDIA releases Open Agent Safety Platform
NVIDIA's platform for autonomous agents uses the OpenShell runtime to isolate agents, restrict what they can access and record their actions instead of relying on the model to police itself.
AI agents are getting powerful enough that NVIDIA is building a security layer to stop them. That tells you where AI is heading. NVIDIA just launched its Open Agent Safety Platform, designed to put enforceable boundaries around autonomous agents as they access files, credentials, networks and real-world systems. The key idea: do not rely on the AI to police itself. Its OpenShell runtime isolates agents, restricts what they can access and records their actions, while additional safeguards can monitor and contain behavior outside the agent itself. Why now? Because AI is no longer only generating text. It is taking actions. Using tools. Navigating systems. Making decisions. And once intelligence can act, the question becomes bigger than how capable the model is: Who sets the boundaries? Who can verify what it did? And who stays in control? That is exactly why Action Model was built around actions, not just outputs. We believe the next generation of AI should be powerful enough to act, but transparent enough to verify, accountable to the people it serves, and ultimately owned by the community contributing to it. The future of AI is not just autonomous intelligence. It is accountable intelligence.