CoreWeave has announced the launch of its CoreWeave Zero Trust AI Framework, a set of security controls designed to let customers oversee access to their own AI models, data and runtime environments.
The new framework was revealed at Nvidia GTC Berlin.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
The company states that this customer-controlled model will allow organisations to use their own security systems with CoreWeave Cloud, instead of adapting to methods mandated by a cloud provider.
CoreWeave says this model is intended to address changing cloud security requirements that have emerged due to the rise of agentic AI. This approach refers to agents that interact with external systems and process sensitive data through various environments and formats.
According to CoreWeave, running AI workloads in the cloud has generally involved trusting the provider with key aspects of data security. This includes the management of encryption keys, privileged access to servers, and control over execution environments.
CoreWeave claims its new framework changes this dynamic by minimising its position within the customer’s security boundary.
CoreWeave product and engineering executive vice president Chen Goldberg said: “Security for AI workloads cannot rest on a provider’s promises about its own behaviour.
“Our customers are running their most valuable models against their most sensitive data. They should not have to take our word for how we handle either.
“The Zero Trust AI Framework replaces assurances with architecture, so the controls that matter stay in the customer’s hands.”
The framework includes six layers of protection. These comprise Remote Key Encryption, which permits customers to manage their own encryption keys without handing authority to CoreWeave, and Support Access Management, where provider access is limited to explicit authorisation.
It also includes infrastructure attestation services that validate computing environments, and support for Nvidia’s Confidential Computing to protect data while in use.
The company says it does not use customer data to train or improve its own models, and that its network incorporates encryption technology to secure data transfers.
According to CoreWeave, these protections are designed to give customers architectural control over their AI security, aiming to reduce reliance on provider assurances.
The company has indicated plans to continue expanding confidential computing support, with specific Nvidia hardware expected to become available on CoreWeave Cloud from the fourth quarter of 2026.
In July 2026, Flow Traders selected CoreWeave as its main cloud platform provider to facilitate AI model training for quantitative trading.
