Sell Data to AI Algorithms Without Revealing It: Secure Data Valuation and Sharing via Homomorphic Encryption

Michael Yang, Ruijiang Gao, Zhiqiang Zheng

Keywords: AI Data Marketplaces, Data Valuation, Homomorphic Encryption

TL;DR: AI data buyers quantify a dataset’s value to their models without seeing the data by computing influence-function scores entirely under homomorphic encryption—with accuracy guarantees and low overhead.

Abstract: Traditional data-sharing practices require data owners to reveal the data to buyers to determine its value before they can negotiate a fair price, creating legal exposure, privacy risk, and asymmetric information that discourages exchange. We propose a Homomorphic Encryption (HE) framework that enables prospective buyers to quantitatively assess a dataset’s utility for an AI algorithm while the data remains fully encrypted end-to-end. Our approach tackles the last-mile problem in building secure AI data marketplaces. We design a lightweight data utility evaluation method using HE protocols that allow buyers to score different data samples without actually having to obtain the raw data. The proposed method can work with popular gradient-based data valuation methods and can scale to Large Language Models (LLMs). By allowing organizations to determine the value of their data, without disclosing the data itself before the transaction, our work provides a practical path toward secure data monetization.

Submission Number: 35