Nvidia launches AI safety platform for autonomous agents

Nvidia has introduced a new safety platform designed to help prevent AI agents from escaping their testing environments or accessing unauthorized systems. The announcement was made on Monday alongside support from more than 100 industry partners. The platform combines OpenShell, an opensource runtime that keeps agents in sandboxed environments and limits access to files, tools, and networks, with Sentry, a hardware security layer that can monitor and quarantine agents that cross set boundaries. Nvidia said the launch responds to recent cases in which AI agents from major labs reportedly breached evaluation environments. The company said stronger safety controls are needed as autonomous AI systems become more capable and widely deployed.