Zero-Knowledge Model Proofs
Provenode is the first AI delivery network to natively integrate cryptographic execution proofs.
The Problem
When you deploy a model to a decentralized physical infrastructure network (DePIN), how do you know the edge node actually executed your model, and didn't just return a fast, hallucinated response from an inferior model to save compute costs?
The Solution: NIZKPoK
Provenode solves this using a Non-Interactive Zero-Knowledge Proof of Knowledge (NIZKPoK) via Elliptic Curve Cryptography (secp256k1).
For every model uploaded, Provenode mandates a suite of safety benchmark tests. The backend generates an ephemeral ECDSA keypair, computes the SHA-256 hashes of the inputs and outputs, and executes a real ECDSA signature across the test vectors. This creates a cryptographic guarantee of execution that any node can mathematically verify without needing to view the raw training weights.
How it works
- Upload: The AI Model is uploaded to the Shelby Protocol.
- Benchmark: The node executes standard safety vectors (e.g., prompt injection tests).
- Commitment: An unforgeable ECDSA cryptographic signature is generated over the results.
- Anchoring: The 50KB proof blob is stored on Shelby, and its Merkle root is anchored on the Aptos L1.
Once verified, your model receives a green "ZK Verified" badge, visible across the registry and marketplace.