NVIDIA Ties Security Directly into AI Infrastructure with New Open Agent Safety Platform
NVIDIA (NVDA) has launched an open-source AI Agent Safety Platform that combines hardware and software security controls. The platform, which includes Sentry running on NVIDIA BlueField DPUs at the silicon layer and OpenShell as a secure agent runtime environment, is designed to ensure the safety of autonomous AI agents.
Security vendors TrendAI and Bedrock Data have announced integrations into the platform, with initial enterprise clients already onboarding the tools. This move supports the bullish view that NVIDIA's business is building an AI infrastructure stack that is hard to swap out.
The bear case, however, points out that OpenShell is open-source and compatible with third-party compute, leaving room for customers to pair agent safety with rival chips if economics or regulation push in that direction. The success of the platform will depend on how many large enterprises adopt it and keep it tied to NVIDIA infrastructure.
NVIDIA's Open Agent Safety Platform also has implications for its hardware and software business, as it ties security directly into where its products already operate within customers' AI workflows.