Protecting Model Weights and Infrastructure

Model weights are a high-value asset that can be stolen or tampered with, and federated learning plus supply chain dependencies create additional exposure points. Confidential computing is a hardware-based defense that protects model weights and training data even from privileged infrastructure operators, directly addressing risks raised by federated learning and supply chain attacks.