[NVIDIA launches open AI agent security platform, covering from testing to deployment] On September 28, NVIDIA announced the launch of an open AI agent security platform, aiming to cover the security management of AI agents from testing to deployment. The platform includes open source software OpenShell and reference system design Sentry. OpenShell is responsible for setting the AI ​​agent operating boundaries, tracking operations and executing security policies. Sentry continuously monitors agent behavior based on NVIDIA BlueField-4 DPU. If out-of-bounds behavior is discovered, it can be isolated and stopped at the millisecond level. NVIDIA said that currently more than 100 institutions are participating in relevant technical cooperation.