A perturb-then-project mechanism that adds Gaussian noise calibrated to global sensitivity of the chosen score release regime and then projects the result onto the feasibility set of valid cosine objects, and evaluates the mechanism across RAG, face recognition, semantic retrieval, image similarity, and recommender-system tasks.
Abstract
A growing number of applications, such as biometrics and retrieval-augmented generation (RAG), rely on cosine similarity scores computed between vector embeddings of text, images, or audio. These systems return similarity scores through their APIs for ranking and verification. However, such releases can leak information about individual records and enable membership inference attacks. While differential privacy (DP) provides a principled metric for quantifying attack risks, na\"ive application of DP mechanisms---such as adding i.i.d. Gaussian noise to vector entries---leads to excessive distortion (i.e., low utility) at a given privacy constraint that scales poorly with the number of released scores. We propose \textsc{ScoreShield}, a perturb-then-project mechanism that adds Gaussian noise calibrated to global sensitivity of the chosen score release regime and then projects the result onto the feasibility set of valid cosine objects. \textsc{ScoreShield} satisfies $(\varepsilon,\delta)$-DP for releasing similarity score vectors and Gram matrices. We provide utility guarantees for the exact Frobenius metric projection used in the risk analysis, and prove convergence to feasibility for the practical averaged alternating-projection solver used for large-scale Gram releases. For full pairwise cosine Gram release under record-level replacement adjacency, the exact-projection bound improves the $n$-dependence of squared Frobenius risk from $\Theta(n^3)$ for the na\"ive Gaussian baseline to $\mathcal{O}(n^2)$ for fixed privacy parameters, with sharper local bounds at low-rank Grams. We evaluate the mechanism across RAG, face recognition, semantic retrieval, image similarity, and recommender-system tasks.
This paper studies embedding-space privacy as a representation-level learning problem. Rather than altering raw records directly, the proposed framework applies embeddingspace transformation to full-record representations through Gaussian perturbation and adversarial representation sanitization. The method is evaluated through ablation across utility metrics, linkage attacks, attribute-inference attacks, and membership-inference tests. The primary empirical evaluation uses a synthetic fusion recommendation benchmark built from MovieLens [1], [2] 32M behavior and Adult-derived demographics [3], while a secondary synthetic medical benchmark is used to examine cross-domain transferability under more constrained conditions. The strongest results appear in the recommendation experiments. Under grouped demographic privacy evaluation, the combined condition preserves recommendation utility with $N D C G {@} K=0.6312$ while reducing exact and entity linkage from 0.7090/0.7204 to 0.0001/0.0000. Sensitive-target attacker performance remains near the majority baseline, supporting the claim of empirical privacy improvement without visible ranking degradation in that benchmark. The healthcare experiments also demonstrate meaningful embedding transformation and linkage reduction, though the current benchmark remains datalimited and therefore less conclusive for utility-focused evaluation. Overall, the findings support the conclusion that embeddingspace transformation can preserve downstream utility while substantially reducing linkage risk and sensitive-information recoverability under explicit attacker evaluation. The findings support embedding-space transformation as a practical privacypreserving strategy for embedding-driven AI systems under explicit attacker evaluation.
D. Panagoulias, Evangelia-Aikaterini Tsichrintzi, E. Sakkopoulos· International Conference on...· 0 citations
LLM applications deployed at scale face a fundamental challenge: privacy constraints prevent direct inspection of user interactions, making it difficult to obtain any representative evaluation dataset or to track the ongoing evolution of production traffic. We present ProxyDrift, a framework that (i) identifies and measures drift between production traffic and offline evaluation sets, and (ii) constructs and refreshes those evaluation sets accordingly; all without access to raw user data. Our approach operates entirely on non-PII proxy representations: structured, multi-dimensional descriptors derived from LLM-based classification of user interactions. We introduce (1) a chance-calibrated, redundancy-aware (RA) alignment score that aggregates per-dimension drift measurements via mutual information; (2) a conditional sampler that generates synthetic proxies respecting inter-dimensional dependencies; (3) a roundtrip consistency analysis that exposes generator/classifier disagreements and guides proxy taxonomy refinement; and (4) a feedback-linkage analysis that ties per-dimension and per-value proxy distributions to user satisfaction, surfacing actionable failure and success modes. Serving hundreds of millions of users, ProxyDrift enables continuous drift monitoring and targeted synthetic data generation without exposing sensitive user data. Experiments confirm strong roundtrip consistency, discriminator-level indistinguishability of synthetic queries from human queries, and tight end-to-end alignment (RA~0.9) with production.
Michael Levit, Josh Ledgard, Haoyu Dong et al.· 0 citations
Vector search powers semantic search, recommendation systems, and retrieval-augmented generation (RAG). By design, the service answering a query sees both the query embedding and, usually, the corpus against which it is matched. This is a privacy breach for both the user issuing the query and the owner of the corpus. A family of cryptographic schemes (e.g., SAP, EMVP, BNTM, Tip-toe) addresses that leak. However, as each scheme is published and evaluated on its own corpus, threat model, parameter choices, hardware, and metric conventions, the numbers cannot be compared directly. Consequently, a practitioner asking which one to deploy today has no defensible way to choose. We close that gap with a uniform experimental comparison, including a Plaintext baseline and four cryptographic backends running over the same workload, hardware, and metric definitions. Under that ruler, the schemes spread across a Pareto frontier in privacy, performance, and recall rather than imposing a flat penalty on performance. We find that the performance of SAP matches Plaintext, EMVP delivers cryptographic indistinguishability at a 4x throughput cost on CPU, BNTM adds malicious-server verifiability at a further 22x median-latency cost, and Tiptoe hides the cluster choice itself, but incurs a 190x per-query cost compared to Plaintext. GPU acceleration pays off for Plaintext and SAP but not for EMVP or BNTM. All our experiment artifacts are publicly available for reproducibility
Anne-Marie Kermarrec, Rafael Pires, Mathis Randl et al.· 1 citation
Retrieval-Augmented Generation (RAG) enhances large language models by incorporating external knowledge, but existing pipelines typically operate on plaintext data, raising significant privacy concerns. Prior work on privacy-preserving retrieval leverages cryptographic techniques such as homomorphic encryption (HE) and private information retrieval (PIR), but often relies on interactive protocols or ranking-based selection mechanisms that incur high latency and potential information leakage. In this paper, we propose a practical non-interactive encrypted retrieval framework for RAG based on threshold selection. Instead of performing expensive top-$k$ ranking under encryption, our approach selects documents whose similarity scores exceed a predefined threshold, reducing computational complexity from quadratic to linear in the corpus size. We implement this design using CKKS-based homomorphic computation, enabling fully encrypted similarity evaluation and document selection without revealing query content, intermediate scores, or selected indices. To bridge the gap between approximate encrypted computation and discrete token reconstruction, we introduce a precision-stable mask polarization method that ensures accurate recovery of selected documents. Experiments on standard retrieval benchmarks demonstrate that our approach achieves competitive retrieval effectiveness while significantly reducing latency compared to ranking-based encrypted methods. These results highlight threshold-based selection as a practical foundation for scalable and secure RAG systems.
Yang Gao, Gang Quan, Scott Piersall et al.· 0 citations
A comparative study of four audits applicable to pre-trained, black-box face generators, which consistently reveal substantial identity distinguishability while reporting markedly different epsilon estimates that reflect each method's distinct assumptions and finite-sample treatment.
Arman Zareian Jahromi, Vishnu Bondalakunta, M. Shah et al.· 0 citations
Vector embeddings of raw records, while not human-readable, do not preserve record privacy: an adversary can reconstruct training records from a released model even when that model is a simple convex classifier. Differential privacy (DP) is the principled defense, but its noise is calibrated to worst-case indistinguishability, hiding arbitrary single-record substitutions, including those far outside the set of plausible alternatives relevant to a reconstruction adversary. The result is noise far larger than what reconstruction robustness requires, degrading accuracy without a corresponding security benefit. We propose Ball-DP: enforcing epsilon-delta indistinguishability over single-record substitutions restricted to a ball of radius r under a distance metric d in the embedding space. A deployment facing only local reconstruction threats can choose a small r, thereby reducing noise and recovering accuracy. The radius makes the scope of the privacy claim explicit against reconstruction attacks; standard DP is recovered when r covers the entire admissible record domain. We provide noise calibrations for regularized convex learning problems under Ball-DP, and derive corresponding reconstruction-robustness certificates, called Ball-ReRo, that upper-bound an attacker's reconstruction success. By deriving the optimal finite-prior MAP reconstruction attack, we empirically audit Ball-ReRo certificates on seven benchmark learning tasks. Our experiments show that calibrating noise to Ball-DP improves utility, considerably exceeding the dilution of reconstruction robustness in high-privacy regimes, i.e., when epsilon is small.