VAST Data Launches DataEnclave to Enable Confidential AI on Sensitive Enterprise Data

VAST DataEnclave Addresses Critical Enterprise AI Challenge

VAST Data announced VAST DataEnclave, a confidential AI capability within the VAST DataEngine that enables organizations to run advanced AI models on sensitive data without compromising either the data or the model's intellectual property3

3

. The platform targets a fundamental problem in enterprise AI: sensitive data often cannot move to where leading AI models run, while model providers are reluctant to distribute proprietary model weights into infrastructure they don't control4

4

.

Financial institutions, healthcare providers, and government agencies hold valuable data that was never intended to leave their controlled environments due to sovereignty rules, regulations, and security requirements2

2

. According to Phil Manez, VP of strategic initiatives at VAST Data, the opportunity cost of not having advanced AI models access this data is reaching a critical point2

2

.

Hardware-Isolated Environments and Cryptographic Attestation Protect Both Sides

VAST DataEnclave uses NVIDIA Confidential Computing to create hardware-isolated environments where proprietary AI models and enterprise data can be decrypted and processed securely3

3

. The system employs trusted execution environments spanning CPUs and GPUs, encrypting guest memory, GPU memory, and NVLink traffic while isolating active models and data from administrators, infrastructure operators, and other tenants3

3

.

Before any workload begins, cryptographic attestation verifies the hardware, firmware, runtime environment, and workload policy1

1

. Sensitive information such as encryption keys is released only when the environment matches the approved configuration1

1

. This "verify before decrypting" principle ensures that both customer data keys and model weights remain under their respective owners' control throughout processing3

3

.

Protecting Data During Processing Becomes Critical for AI Workloads

While organizations have long focused on protecting data at rest and in transit, AI introduces a new challenge: protecting data during processing1

1

. Confidential computing addresses this by isolating working memory from the host operating system, hypervisor, and administrators with root access2

2

. For AI workloads, this protection extends to GPUs, where training and inference put model weights, prompts, and intermediate results into memory during use2

2

.

The secure runtime ensures that a model running inside the enclave never accesses proprietary records directly. Instead, a retrieval application outside the enclave searches enterprise documents, assembles relevant excerpts into a prompt, and sends that prompt to the model's inference API, making the prompt the only enterprise data the model ever sees2

2

.

Model Weights Represent Billions in Investment and Intellectual Property

For model builders, weights represent billions of dollars of training investment and valuable intellectual property3

3

. Jeff Denworth, VAST Data's co-founder, stated that "model weights are fast becoming the most valuable intellectual property in the world"3

3

. Shipping these weights into a customer's data center puts a company's core asset inside someone else's building, where anyone who can read GPU memory could copy the weights and stand up the model elsewhere2

2

.

DataEnclave's attestation server belongs to the model owner, and every launch, including each new replica, must pass verification again2

2

. This gives model builders an audit trail showing their weights were decrypted only in environments they cleared, maintaining control even after deployment2

2

.

Full-Stack Approach Integrates Compute, Networking, and Security

Cisco approaches confidential AI as a full-stack infrastructure challenge, recognizing that secure execution alone isn't sufficient for production AI1

1

. The architecture requires scalable AI compute, high-performance networking to move data between compute, storage, and accelerators, and intelligent data infrastructure from VAST that provides a scalable platform for AI workloads1

1

.

Cisco AI Defense addresses security risks across AI models and applications, while Splunk and Isovalent provide observability across applications, infrastructure, networks, and cloud-native environments1

1

. Combined with auditable records of attestation, policy decisions, and workload activity, organizations can monitor operations without breaking the isolation that confidential AI depends on1

1

.

Deployment Flexibility and Partner Ecosystem Support Adoption

VAST DataEnclave can be deployed in customer data centers, AI clouds, and fully air-gapped environments3

3

. The system logs attestation events, key releases, and enclave lifecycle activity in the VAST DataBase data warehouse, providing an audit trail without revealing protected data or model weights3

3

.

Initial partners include AI model builders like Cohere, CrowdStrike, Deepgram, The San Francisco AI Factory, Fundamental Research Labs, and TwelveLabs3

3

. Hardware partners Cisco and Super Micro Computer will offer integrated systems3

3

. VAST includes the capability in the existing AI Operating System rather than charging separately, with DataEnclave scheduled to ship in the first quarter of 20273

3

.