Security News

New 'Confidential AI' Tool Lets Businesses Run AI Models Without Exposing Sensitive Data

Australian Cyber Security Magazine · 23 Sept 2026

Key Takeaway Businesses in regulated industries should look for AI platforms that verify security controls before data ever leaves customer-controlled infrastructure, rather than relying on trust alone.

VAST Data has released DataEnclave, a new capability designed to solve a common problem for regulated industries: needing to use AI models on sensitive data without sending that data to an external provider, or exposing a model builder's proprietary model to infrastructure it does not trust. This matters particularly for sectors like financial services, healthcare and government, where data movement restrictions are strict.

The system works by using hardware-isolated execution and cryptographic verification to confirm an environment is trustworthy before any sensitive data or model assets are decrypted and loaded. Customers keep control of their own data encryption keys, while model owners retain control of their model's keys and weights, limiting what infrastructure operators or administrators can access during processing. Australian provider Sharon AI said it plans to use the capability to offer onshore hosting where both model and customer data remain protected from third parties, including the provider itself.

VAST frames DataEnclave as part of a broader strategy to manage AI models as governed resources, similar to how data is governed, controlling which models can run, on what data, and under what conditions. The technology is built on NVIDIA's confidential computing framework, encrypts memory and GPU traffic, and logs attestation and lifecycle events for audit purposes. It supports both connected and air-gapped deployments.

confidential computing AI security data sovereignty

Summarised by CISO AI from Australian Cyber Security Magazine. We link back to every original so you can read it yourself.