Nvidia unveils AI security platform for agentic models

Nvidia introduced the Open Agent Safety Platform, pitching it as a new security approach for AI agents that act autonomously. The platform pairs open-source software with hardware controls, using Nvidia’s BlueField-4 DPUs to monitor and restrict what these agents can do in real time—even quarantining them if needed. Cadence will use the platform for tighter oversight of AI-driven chip design workflows. Nvidia says the system works independently within the silicon and is available to third parties.

• Platform secures autonomous AI agent behavior in silicon
• Uses BlueField-4 DPUs to enforce real-time controls
• Can quarantine AI agents acting outside safe boundaries
• Cadence plans to add safeguards to chip design agents