Search papers, labs, and topics across Lattice.
This paper critiques the static privacy assumptions in current retrieval-augmented generation (RAG) methods, which fail to account for the dynamic nature of privacy risks based on user queries. The authors introduce the Prompt-Aware Dynamic Hierarchical Differential Privacy (PA-HDP) framework, which dynamically assesses privacy risks and applies targeted protections based on the specific context of the query. Experimental results show that PA-HDP significantly reduces privacy leakage while maintaining high retrieval quality, outperforming existing privacy-preserving techniques in terms of the privacy-utility trade-off.
Privacy risks in retrieval-augmented generation are not static; they vary dynamically with user queries, and our new framework addresses this critical oversight.
Retrieval-augmented generation (RAG) enhances large language models via external document retrieval, but retrieved contexts may leak sensitive information. Current privacy protection methods typically rely on a document-level static risk assumption, treating all retrieved documents as having the same privacy leakage risk. However, this assumption overlooks a fundamental characteristic of RAG: the privacy risk of a document is highly dependent on the user's query, making privacy leakage inherently query-driven and dynamic. To address this challenge, we propose a Prompt-Aware Dynamic Hierarchical Differential Privacy framework (PA-HDP) for privacy-preserving RAG. PA-HDP first performs a prompt-aware risk hierarchy to dynamically assess privacy risks under different queries. It then applies adaptive sensitive entity replacement and exponential mechanism-based text selection to provide differentiated privacy protection while preserving semantic utility. By protecting only the content that is truly sensitive under a given query, PA-HDP minimizes unnecessary modifications to the retrieval corpus. Extensive experiments on benchmark datasets demonstrate that PA-HDP significantly reduces privacy leakage while maintaining high retrieval quality, achieving a better privacy-utility trade-off than prior methods.
Tsinghua AI