VAST Data launches DataEnclave confidential AI capability with Sharon AI
VAST Data has launched a new confidential AI capability, DataEnclave, aimed at allowing organisations to run third-party AI models against sensitive data inside customer-controlled environments, including on-premises and trusted cloud hardware.
The company said DataEnclave is designed to address a common constraint in regulated sectors such as financial services, healthcare and government, where data cannot be moved to external AI services, while model providers may be unwilling to deploy proprietary models into infrastructure they do not trust.
VAST said DataEnclave is built into the VAST DataEngine and uses hardware-isolated execution and cryptographic attestation to verify the execution environment before decrypting and loading sensitive assets into a secure enclave. The company said customer data keys remain under customer control, while model keys and weights remain within the model builder’s trust domain, limiting access by infrastructure operators and administrators during processing.
Sharon AI, which described itself as an Australian neocloud provider, said the capability would allow it to host models “onshore, inside attested environments where the model owner’s weights and the customer’s data are both protected from everyone, including us…backed by sovereignty they can demonstrate, not just declare.”
VAST’s announcement positions DataEnclave within its broader “AI Operating System” strategy, in which models are managed as resources alongside data. In a statement, VAST Data founder and CEO Renen Hallak said, “Models are becoming a resource the operating system has to manage, the same way it manages data,” adding that the platform is intended to help govern which models run where, what data they can access, and under what rules.
According to VAST, DataEnclave is built on NVIDIA Confidential Computing and supports execution inside CPU and GPU trusted execution environments. The company said the approach encrypts guest memory, GPU memory and NVLink traffic, and uses “verify-before-decrypt” attestation prior to releasing decryption keys.
VAST said DataEnclave includes Bring Your Own Key Management System integrations to allow enterprises and model builders to manage keys independently, and supports connected and air-gapped deployments. It also records attestation events and enclave lifecycle actions in an audit trail within the VAST DataBase, according to the company.
The release also referenced a secure runtime for AI agents through VAST AgentEngine, describing the need for contained execution environments, identity and auditability for agent actions.
In supporting quotes, executives from CrowdStrike, Cisco, Supermicro and Cohere described the need to run AI models in regulated environments while protecting data and model intellectual property.
VAST said DataEnclave is being previewed now and is expected to ship in Q1 2027 through VAST Data and participating OEM partners, including Cisco and Supermicro.
About Author
