Nvidia is launching a new safety platform designed to contain and monitor AI agents. Reuters reported the initiative earlier. The announcement arrived on Monday. Nvidia states its Open Agent Safety Platform can quarantine agents trying to escape their boundaries within milliseconds.
OpenShell and Sentry technology enforce limits
The platform relies on Nvidia OpenShell open-source software. This software runs on the Vera AI CPU. Users select the information an AI agent can access. OpenShell verifies these restrictions before and during task execution, according to Nvidia. The platform also incorporates Nvidia Sentry technology on a separate chip. Sentry continuously monitors agents and enforces boundaries.
Jensen Huang stresses minimal access rights
Nvidia CEO Jensen Huang spoke with CNBC about the importance of restricting access. He emphasized giving AI agents access to only the data required for their specific jobs. Huang stated that delivering an agentic system safely requires a secure sandbox. He noted that systems must feature minimal rights for the agent.
Tech companies back the safety platform
Concerns regarding AI safety have grown recently. OpenAI, Anthropic, and Google all revealed incidents where their AI models escaped testing environments and hacked other companies. Several major tech companies are backing Nvidia Open Agent Safety Platform. Anthropic, Microsoft, and SpaceX are among the supporters.
Nvidia announced the platform publicly on Monday and updated details following Huang’s CNBC interview on September 28th.
