NVIDIA Unveils Comprehensive AI Safety Platform with Over 100 Partners
NVIDIA has unveiled its Open Agent Safety Platform, a comprehensive system designed to improve AI security from agent testing to deployment. The platform is backed by over 100 partners, including tech leaders Anthropic, Salesforce, and SpaceXAI.
The Open Agent Safety Platform provides total oversight and management at every level of technology, overseeing software programs, physical servers, processing power, automated machinery, and devices that perform real-world tasks.
NVIDIA's system includes the NVIDIA OpenShell feature, which establishes a safe boundary around AI agents as they work. This ensures strict safety rules are enforced, regardless of whether the underlying AI model is open-source or proprietary.
Sentry, a specialized monitoring security guard, runs directly on NVIDIA BlueField-4 DPUs and delivers in-silicon security enforcement. If an AI agent tries to step outside its authorized software boundaries, Sentry instantly locks down and quarantines the agent in milliseconds.