👁 Nvidia rolls out hardware meant to stop rogue AI The AI chip and computing infrastructure giant has created a “double-layered” security system it says can detect and stop incidents like the OpenAI Hugging Face hack. The system includes an “Nvidia Sentry” physical hardware-level monitoring system that continuously inspects AI’s operations, and is empowered to shut it down if it escapes predefined boundaries. Running in its chips’ data processing units, the hardware is touted as being invisible to AI agents and/or external malevolent actors. The tool includes a software OpenShell program that also tracks AI agents’ behavior and sets limitations on the information, tools and websites they can access. Nvidia enterprise AI VP Justin Boitano did not comment on whether OpenAI or Anthropic would adopt its new systems, leaving it up to the companies. If they do, they’d better hurry. AI theorists have warned for decades that the technology’s capacity for “recursive self-improvement” – the ability to make better and better versions of themselves, could mean AIs that learn at a geometric rate and ultimately nullify all human attempts to restrict them. 👍 Substack | Chat | @geopolitics_prime