NVIDIA launches the Open Agent Safety Platform to keep AI agents from breaking out
Category: Agent Safety • Infrastructure
What it is
NVIDIA has released the Open Agent Safety Platform, a software layer to stop AI agents escaping containment, following sandbox-escape disclosures from OpenAI, Anthropic, Meta, and Google. It includes OpenShell, which runs on central processors and sets limits on what agents can do, and Sentry, which monitors agents. The launch rests on the argument that model-level safeguards alone cannot govern what agents can access or do. Named partners include Cisco, Microsoft, Oracle, CoreWeave, Dell, HPE, Lenovo, ARM, and Intel.
Why it Matters for Enterprises
Agent safety is shifting from model behavior to infrastructure controls such as permissions, isolation, and monitoring. Enterprises should ask AI and cloud vendors how agent actions are limited and logged and align these controls with existing security and ERP access policies before agents get write access.
