Research
Shadow Queries for Private Retrieval in Vector Databases
Publication date not yet evaluated · Added arxiv.org
Page published
Coverage timeline
Single-source research — one report is available.
Why it matters
Embedding inversion attacks let adversaries reconstruct sensitive source text from embeddings stored in cloud vector databases, so a defense that breaks the coupling between embeddings and documents directly addresses a real data-exfiltration risk in RAG deployments.
A research paper introduces Shaq (shadow query generation), a defense against embedding inversion attacks (EIAs) on cloud-based vector databases used in RAG systems. Shaq decouples stored embeddings from original documents by synthesizing diverse shadow queries per document and indexing their embeddings in place of the originals, reporting reduced recovery rates while preserving retrieval utility.