Information Leakage From Prices in Query-Based Data Markets

Query-based pricing enables personalized data acquisition for buyers, exhibiting strong potential in data markets. Prior studies design arbitrage-free price functions to accelerate query price computation. However, the proposed prices often correlate with query answers, leaking information about the underlying data. In this paper, we study information leakage and attack surfaces in query pricing with the goal of designing mitigation strategies that mislead attackers at the pricing stage, which maintains bounded overpayment for buyers. We introduce three price-only inference attacks that derive sensitive information at no financial cost, when investigating how query prices leak information about data existence and cardinality. To reduce the leakage risk, we propose PerPricer, a perturbationbased pricing framework that releases masked prices computed on perturbed data with bounded overpayment guarantees for buyers. It is suitable for existing query pricing mechanisms. Extensive experiments on real-world datasets and benchmarks verify the existence of information leakage in query pricing and the effectiveness of the proposed PerPricer.