The Cryptocurrency Post The CryptocurrencyPost
AboutContact
Technology

NEAR Protocol Confirms Verifiable Private Inference for AI

Photorealistic data center enclave glowing with hardware attestation and network nodes for private inference.

NEAR Protocol has detailed a new technical approach to AI execution, confirming that NEAR AI now utilizes secure hardware enclaves to provide verifiable private inference. The system is designed to return hardware-signed proofs that verify the specific model used, the data processed, and the execution itself, addressing growing concerns over data sovereignty and the limitations of closed AI models.

The development shifts the trust model from contractual agreements to cryptographic and hardware-level certainty. By running AI agents within a user-owned stack, NEAR aims to provide a structural alternative to centralized AI providers, particularly in light of increasing export controls and data privacy restrictions.

Secure Enclaves and Hardware Proofs

At the core of this update is the use of Trusted Execution Environments (TEEs), such as Intel TDX and confidential GPUs. According to official NEAR AI documentation, these secure enclaves allow inference to run in an isolated environment where memory is encrypted at the CPU level. This prevents host operators, hypervisors, or unauthorized third parties from accessing the data being processed.

The system generates a cryptographic “attestation” or hardware-signed certificate. This proof allows users or third parties to verify that the workload ran exactly as intended without being modified. The NEAR Protocol official account noted that the IronClaw security layer is used to protect the agent level, ensuring that users maintain sovereignty over their data and model interactions.

Addressing Data Sovereignty

The move toward verifiable inference comes as a response to the “closed” nature of frontier AI models. In typical cloud-based AI interactions, users must rely on the provider’s contractual promise that data is not being stored or used for training. NEAR’s implementation replaces this reliance on trust with “structural assurances,” where the silicon itself proves the security of the environment.

This approach is particularly relevant for:

  • Export Controls: Providing verifiable proof of hardware and execution locations.
  • Sensitive Workloads: Allowing institutions to run models on rented cloud compute without exposing proprietary data to the cloud provider.
  • Model Integrity: Ensuring that the specific version of an AI model requested is the one actually performing the task.

Status and Integration

While the technical framework for private inference and hardware attestation is now officially documented and confirmed, specific adoption metrics remain pending. The available sources do not yet provide data on total usage numbers or a comprehensive list of third-party integrations launched within the last 48 hours. The current focus remains on the deployment of human-owned AI stacks that leverage these secure hardware proofs to bypass centralized bottlenecks.