VAST Data Launched Confidential AI Runtime System
The new DataEnclave platform uses hardware attestation to secure model weights and sensitive data during AI inference.
Updated on Sept. 26, 2026 in Artificial Intelligence
Live Poll
Do you trust that new security tools can sufficiently protect your sensitive data from AI model leaks?
VAST Data released DataEnclave, an AI runtime system integrated into the VAST AI Operating System that secures sensitive data and model weights within hardware-isolated environments. The platform allows enterprises to maintain control of encryption keys while enabling AI processing in both connected and air-gapped deployments.
Why it matters
The system addresses security vulnerabilities that occur when AI models must decrypt sensitive data within GPU memory during inference. By shifting these processes to a hardware-attested environment, the platform aims to provide a more secure execution layer for enterprise AI agents.
DataEnclave utilizes Nvidia Confidential Computing to verify the integrity of the computing environment before granting access to sensitive model weights. The system logs all lifecycle actions and attestation events in a tamper-proof audit trail to ensure transparency.
The players
VAST Data
An enterprise data infrastructure provider specializing in AI-ready storage and computing software platforms.
Nvidia
A hardware manufacturer and software ecosystem leader providing the GPUs and Confidential Computing architecture that power modern AI workloads.
Cohere
An AI research firm focused on building large language models for enterprise-specific applications.
CrowdStrike
A cybersecurity firm providing cloud-delivered protection and threat intelligence services.
Cisco
A global networking and enterprise server hardware provider that serves as a launch partner for the system.
The details
DataEnclave works by isolating the execution of AI agents through VAST's AgentEngine, a system designed to secure workflows by decrypting data only after the computing environment has been hardware-verified. By utilizing hardware-isolated execution, the system prevents unauthorized access to model weights while they are active in GPU memory. This process relies on independent key control, ensuring that the enterprise, rather than the infrastructure provider, retains authority over encryption keys throughout the AI lifecycle.
Timeline
September 26, 2026: VAST Data officially launched the DataEnclave system.
The Tech Race
The platform enters a crowded space where infrastructure providers are racing to secure AI models against memory-based attacks. It follows the industry-wide effort to move beyond simple encryption-at-rest toward verified, hardware-isolated processing for enterprise LLMs.
Enterprises can begin integrating this runtime into their existing VAST infrastructure, provided they use compatible servers from partners like Cisco, Lenovo, or Supermicro. The system is designed to support both standard network-connected operations and highly secure, air-gapped environments.
The takeaway
This launch signals that security-conscious enterprises now require hardware-level verification for AI inference workloads rather than relying solely on traditional software perimeters. Interested users should watch for the company's upcoming performance benchmarks to determine the latency cost of these security protections.
What happens next
VAST Data has committed to publishing future benchmark reports that will detail the performance overhead associated with the system's hardware-attestation processes.
Further reading
For more on the current state of infrastructure for intelligent applications, browse our latest coverage on Artificial Intelligence.
Source note: This article includes information reported by The Manila times.
Live Poll
Do you trust that new security tools can sufficiently protect your sensitive data from AI model leaks?
