Nvidia Unveils New Safety Platform to Keep AI Agents Under Control

Nvidia has announced a new software platform designed to improve the safety and control of AI agents. Called the Open Agent Safety Platform, the system is built around two key components — OpenShell and Sentry. OpenShell is intended to define the permissions available to an AI agent and prevent it from exceeding those limits. Sentry, meanwhile, functions as an independent monitoring layer that continuously observes agent activity and can intervene when suspicious behaviour is detected. Nvidia executive Justin Boitano said OpenShell controls an agent’s actions, while Sentry independently monitors and blocks potentially unsafe behaviour.

Nvidia said more than 100 organisations, including Microsoft, Perplexity and JPMorgan Chase, were already using the platform at the time of its launch. The announcement comes amid growing discussions around the need for stronger safeguards as AI systems become increasingly capable of performing tasks autonomously. Separately, OpenAI has reportedly paused the release of its Astra 6.1 AI model after internal testing indicated that it did not meet the company’s required safety standards.