Nvidia launches open AI agent safety platform, with Anthropic and others on board

Nvidia has launched the NVIDIA Open Agent Safety Platform, which provides full-stack governance and control for AI agents, agent hardware, and robotic systems. The platform includes OpenShell security software, which sets boundaries for AI agents running on CPUs and supports both open-source and closed-source models.

OpenShell uses very little performance on the Vera platform. When combined with Vera, it lets AI agents finish work quickly while staying safe. As open-source software, it also supports third-party platforms including Arm and Intel products.

Nvidia also announced NVIDIA Sentry, a watchdog system that continuously monitors AI agent behavior. If an AI agent tries to break through software limits, Sentry can isolate and stop it within milliseconds. Sentry runs on BlueField-4 DPU and enforces security policies independently at the chip level.

Anthropic, SpaceX, and Scale AI have already partnered with Nvidia to use OpenShell for their AI models and agents.