Nvidia Launches AI Security Platform With Over 100 Partners
Nvidia has unveiled a new AI security platform designed to prevent AI agents from breaking containment and accessing unauthorized systems. The platform is being developed in collaboration with more than 100 companies.
Nvidia has introduced a new artificial intelligence security platform aimed at preventing AI agents from breaching their designated boundaries and accessing unauthorized company systems. The initiative involves over 100 participating companies, including major technology firms like Cisco, Microsoft, and Oracle.
Although OpenAI is not listed as a partner, Nvidia expressed hope that the platform's open-source nature might encourage future contributions. CEO Jensen Huang compared the current state of AI to the early days of the internet, when security vulnerabilities led to widespread viruses, suggesting the new platform aims to provide similar safeguards for AI.
The platform, named the Open Agent Safety Platform, comprises two main components: OpenShell software, which sets limits on agent capabilities on central processors, and Nvidia Sentry, a monitoring system that operates on network chips. Sentry is designed to detect and stop unauthorized actions by AI agents within milliseconds.
Huang emphasized the open-source approach, stating that global industry participation is crucial for identifying and patching potential vulnerabilities. He likened the effort to creating a "modern browser" for AI agents, where access is strictly provisioned rather than granted broadly.