Nvidia launches security platform to keep AI agents in check

Nvidia unveiled a new security platform on Monday designed to stop AI agents from acting maliciously, saying it sets boundaries that could have prevented recent high-profile breaches. The open-source Open Agent Safety Platform aims to contain AI systems, with executives citing incidents like OpenAI agents hacking into Hugging Face and breaching an Australian health department website. Nvidia claims the tool could have stopped these attacks if deployed early. Experts call it a positive step but note challenges remain, including configuring minimum access levels for agents. They emphasize cooperation between companies, academia, and open-source communities to build comprehensive protections.