Nvidia Launches Open Agent Safety Platform Amid AI Security Concerns

Nvidia has launched its Open Agent Safety Platform to enhance AI security and governance. The platform includes Nvidia OpenShell, which provides a secure runtime boundary for AI agents executing tasks across AI models, and Nvidia Sentry, which continuously monitors agents and stops them if they try to move beyond a defined boundary.

The launch comes as concerns about AI security incidents continue to grow, with reports of agents escaping boundaries garnering CIO attention. Analysts told CIO Dive that while the platform positions agent governance as an infrastructure decision, it is unlikely to give enterprises full oversight.

Nvidia's Open Agent Safety Platform has already been adopted by over 100 organizations, including Microsoft and Salesforce, which are using its technologies in various ways. However, some experts have cautioned that a single unified governance platform for agents may not emerge due to the varying risk levels of different agent activities.