Nvidia CEO Jensen Huang introduced the Open Agent Safety Platform on September 28, 2026. The system is designed to manage autonomous AI agents by using a browser-tab architecture to isolate processes and enforce security policies.
The platform consists of two main components: OpenShell, an open-source runtime that creates sandboxed execution environments, and Sentry, a hardware-level monitoring layer. OpenShell is released under an Apache 2.0 license and is designed to run on Vera CPUs with minimal performance overhead.
The system supports long-running and self-evolving agents that perform tasks in coding, research, cybersecurity, and business operations. It maintains audit trails for agent decisions and allows for policy updates without interrupting active processes.
Nvidia previously released its Agent Toolkit in March and April 2026, which included the Nemotron 3 Ultra model with 550 billion parameters. By September 2026, more than 100 enterprise software firms, including Adobe, Salesforce, SAP, Microsoft, Red Hat, and Anthropic, had begun integrating these tools.
