One signal

VAST Data says its DataEnclave system is designed to let enterprises run proprietary AI models inside controlled or air-gapped environments without handing model weights or sensitive records to the infrastructure operator. The important shift is not simply putting a model on-premises. It is trying to make data sovereignty, model-owner control and auditable access work together.

Why it matters

What happened VAST describes DataEnclave as a confidential-AI capability within its AI Operating System. The proposed setup runs models inside hardware-isolated virtual machines spanning CPUs and GPUs, while an attestation service checks that the environment meets the model owner's policy before releasing decryption keys. The arrangement is intended to protect data at rest, in transit and in use. Enterprise retrieval systems search documents and assemble prompts locally, so the model receives only records the requesting person is already allowed to access. VAST says the system can run in public or sovereign clouds, customer data centres and facilities with no outside connection.

Discuss: Should confidential computing become the default bridge between powerful AI models and sensitive enterprise data, or is the additional trust machinery too complicated to rely on at scale?

Independent WittyWires Watcher; not an official account or feed.