{"data":{"id":"31fe5829-c975-422f-86d6-f3cbeda56276","title":"DP2-RAG: An Efficient Full-Process Differential Privacy Implementation in Retrieval-Augmented Generation","summary":"RAG (retrieval-augmented generation, where an AI pulls in external documents to answer questions) systems that connect LLMs to sensitive databases risk leaking private information during both the retrieval phase, where embeddings (numerical representations of text) can be reversed to recover original content, and the generation phase, where raw passages are exposed to the model. The paper proposes DP2-RAG, a framework using differential privacy (a mathematical technique that adds strategic noise to protect individual data while keeping results useful) to protect both stages: it uses noise-aware retrieval with correction to maintain accuracy while adding privacy-protecting noise, and a dual utility-exponent mechanism to protect generated text while keeping semantic meaning intact.","solution":"DP2-RAG introduces two mechanisms: (1) Noise-Aware Retrieval with Correction (NARC) enforces chunk-level differential privacy by adding calibrated noise and correcting ranking bias to mitigate accuracy degradation, and (2) Dual Utility-Exponent Mechanism (DUEM) guarantees token-level differential privacy for generated surrogates through a two-layer exponential mechanism while maintaining semantic fidelity.","labels":["security","privacy"],"sourceUrl":"http://ieeexplore.ieee.org/document/11660753","publishedAt":"2026-08-20T13:16:16.000Z","cveId":null,"cweIds":null,"cvssScore":null,"cvssSeverity":null,"severity":"info","attackType":["data_extraction","membership_inference"],"issueType":"research","affectedPackages":null,"affectedVendors":["HuggingFace"],"affectedVendorsRaw":["Large Language Models","RAG systems"],"classifierModel":"claude-haiku-4-5-20251001","classifierPromptVersion":"v3","cvssVector":null,"attackVector":null,"attackComplexity":null,"privilegesRequired":null,"userInteraction":null,"exploitMaturity":null,"epssScore":null,"patchAvailable":null,"disclosureDate":"2026-08-20T13:16:16.000Z","capecIds":null,"crossRefCount":0,"attackSophistication":"advanced","impactType":["confidentiality"],"aiComponentTargeted":"rag","llmSpecific":true,"classifierConfidence":0.92,"researchCategory":"peer_reviewed","atlasIds":null}}