CoreWeave launches six-layer security framework for AI workloads

CoreWeave, Inc. (Nasdaq: CRWV) announced the CoreWeave Zero Trust AI Framework, a customer-controlled security model designed to give enterprise customers authority over access to their AI models, data, and runtime environments. The announcement was made at Nvidia GTC Berlin.

The framework consists of six security layers: Zero Key Access, which allows customers to use their own existing encryption key infrastructure; Zero Node Access, which requires customers to explicitly authorize CoreWeave support access for defined operational needs; Zero Trust Infrastructure, which uses hardware-based attestation to verify the infrastructure executing a workload; Zero In-Use Access, which the company intends to extend using Nvidia's third-generation Confidential Computing across CPUs, GPUs, and the Nvidia NVLink fabric; Zero Model Data, under which CoreWeave states it does not use customer prompts, training data, model weights, or outputs to train its own models; and Zero Trust Network Encryption, using IEEE 802.1AE MACsec to encrypt data in transit across its network backbone.

CoreWeave said the Nvidia HGX Rubin NVL8, which supports confidential computing for agentic AI workloads, is scheduled to arrive on CoreWeave Cloud in the fourth quarter of 2026.

"Security for AI workloads cannot rest on a provider's promises about its own behavior," said Chen Goldberg, executive vice president of product and engineering at CoreWeave. "The Zero Trust AI Framework replaces assurances with architecture, so the controls that matter stay in the customer's hands."

Dave McCarthy, group vice president of Cloud and Datacenters at IDC, said the framework "shifts trust from provider promises to technical controls, a critical step for enterprises deploying proprietary AI at scale."

You May Also Be Interested In

• US Treasury: 'Authorizing Wind Down Activities...the Offloading of Cargo Involving Certain Persons or Vessels Blocked on October 21, 2026'
• CoreWeave launches physical AI tool templates for robotics and AV teams
• OpenAI publishes cybersecurity blueprint for AI defenders