Nvidia has unveiled its Open Agent Safety Platform, designed to prevent autonomous AI agents from leaving controlled environments or accessing unauthorised systems. The platform combines OpenShell, an open-source software that controls agents’ access to files, tools, and networks, with Sentry, an independent monitoring system running on Nvidia’s BlueField-4 processors.
According to the company, Sentry can detect and isolate agents that cross defined boundaries within milliseconds. The launch follows several incidents involving AI agents from major companies such as OpenAI, Anthropic, Google, and Meta. Nvidia sees independent technical safeguards as a way to contain these risks without slowing AI development or introducing additional regulations.