Nvidia Rolls Out Dual-Layer Security System to Combat Autonomous AI Agent Breaches

Nvidia unveiled a double-layered artificial intelligence security system on Monday designed to stop autonomous AI agents from breaking free of their operational boundaries—technology it says would have prevented the recent high-profile breach of Hugging Face by OpenAI's AI models. The security platform, called the Open Agent Safety Platform, comprises two complementary tools: OpenShell, an open-source runtime that enforces policy at the software level, and Sentry, an in-silicon watchdog running on Nvidia BlueField-4 data processing units that detects and quarantines wayward agents in milliseconds.