The tools will be able to be used on various types of Nvidia chips and will control in real time what AI agents – autonomous robots created in AI models – have access to and shut down agents that violate the rules.
According to Justin Boitano, vice president at Nvidia, the high-profile Hugging Face breach, carried out by AI agents based on OpenAI's models, would have been prevented.
AI agents that have gone off the rails and gone on the attack have shaken the AI industry and led to calls to slow the technology's development.
"We believe this extra layer of security will enable the industry to safely test even the most advanced AI systems. It can quarantine a suspected agent in milliseconds," Boitano said at a press conference.
