CoreWeave (CRWV) said Wednesday it launched a security framework called Zero Trust AI that keeps the company out of customers' data, and pre-built software templates called Physical AI Launchables for robotics and self-driving vehicle work.
Under the Zero Trust AI Framework, customers keep their own encryption keys and grant CoreWeave access to servers only when needed for a specific job, the company said.
It also ruled out using customer prompts, training data, or model weights to train its own models, and put hardware-based confidential computing on the roadmap for Q4, CoreWeave said.
Separately, the company said Physical AI Launchables ships tested configurations for the tools robotics and self-driving teams use most, so a run starts far quicker.
The templates cover Nvidia (NVDA) Isaac Sim, CARLA, MuJoCo, Nvidia AlpaSim and Nvidia Cosmos 3, plus partnerships with Cadence, PhysicsX and other software vendors, the company said.
CoreWeave said it introduced the Zero Trust AI Framework and Physical AI Launchables at Nvidia's GTC Berlin.
Shares of CoreWeave were up 3% in Wednesday premarket activity.
