CoreWeave Launches Zero Trust AI Framework at NVIDIA GTC
CoreWeave launched a Zero Trust AI Framework in Berlin to give customers direct technical control over their AI models, data, and runtime environments.
CoreWeave, Inc. unveiled the Zero Trust AI Framework at NVIDIA GTC Berlin to shift security control from cloud provider assurances to technical architecture. The framework implements six security layers: Zero Key Access, Zero Node Access, Zero Trust Infrastructure, Zero In-Use Access, Zero Model Data, and Zero Trust Network Encryption. These layers are designed to remove the provider from the customer's trust chain, ensuring customers maintain authority over their models and data.
The framework integrates NVIDIA's third-generation Confidential Computing across CPUs, GPUs, and the NVIDIA NVLink fabric. This includes the NVIDIA HGX Rubin NVL8, which is scheduled for deployment in the CoreWeave Cloud during the fourth quarter of 2026. CoreWeave also confirmed that it does not use customer prompts or training data to improve its own internal AI models.
Chen Goldberg, Executive Vice President of Product and Engineering at CoreWeave, stated that the framework replaces assurances with architecture so that controls remain in the customer's hands. Dave McCarthy of the International Data Corporation described the move as a critical step for enterprises moving sensitive and regulated data into AI workflows, noting that zero trust must extend to the entire infrastructure stack.
