NVIDIA released an AI security framework focusing on the full agent stack to secure AI operations. It is the company's first major update in the safety domain since its initial AI security initiatives.
NVIDIA reported that the framework enforces policies outside the agent’s reach and provides sandboxed execution, ensuring secure access to data, network and system resources. That compares with earlier security measures that lacked such comprehensive controls.
The framework is built on NVIDIA's OpenShell, targeting secure AI agent operations across various applications. Availability begins with Open Secure AI Alliance partners, initially for cybersecurity and infrastructure teams.
"AI security is an engineering problem. That means defined security requirements, enforceable controls, named owners and evidence that protections work," said Jensen Huang, CEO of NVIDIA. The framework aims to address the growing need for secure AI deployment as capabilities advance.
The announcement follows increased industry demand for secure AI deployment. NVIDIA said the framework addresses the growing need for secure AI deployment as capabilities advance.
NVIDIA did not specify exact deployment timelines, and the open question remains on how quickly organizations can adopt the framework. The company said the framework will be further developed with collaboration from the Open Secure AI Alliance.
Source: nvidia
