Antonio Scardace, Francesco Guarnera, Sebastiano Battiato +1cs.CV
While deep generative models offer new opportunities for medical image synthesis and data sharing, their ability to memorize and reproduce training samples raises serious concerns about patient confidentiality. Detecting such memorization at scale remains challenging: traditional pixel-based metrics are sensitive to generation artifacts, whereas generic embedding-based metrics often lack the anatomical sensitivity required for medical data. To address this challenge, we introduce DeepSSIM++, a self-supervised similarity metric for scalable memorization auditing in medical generative models. By leveraging multi-scale feature aggregation and anatomy-preserving augmentations, DeepSSIM++ learns an embedding space where cosine similarity approximates the Structural Similarity Index (SSIM), eliminating the need for exact pixel-level registration. Compared with state-of-the-art baselines, DeepSSIM++ achieves an average Macro F1 improvement of 33 percentage points under ideal alignment and 46 percentage points under realistic spatial and intensity perturbations. Furthermore, it accelerates large-scale similarity computation by several orders of magnitude compared with analytical SSIM. By combining anatomical sensitivity and computational efficiency, DeepSSIM++ provides an open-source tool for scalable memorization auditing in medical generative AI. Code and data are publicly available at: https://github.com/brAIn-science/DeepSSIM.
Differential privacy (DP) has traditionally been used to provide theoretical upper bounds on an algorithm's stability to changing its training data. In modern private machine learning applications, achieving strong tradeoffs between utility and theoretical privacy is challenging, and thus one may optimistically hope that existing theoretical privacy analyses are loose. Recent work on privacy auditing has adopted a dual viewpoint, instead lower bounding the true privacy of an algorithm by constructing empirical distinguishing events. The auditing literature has thus far yielded a pessimistic outlook on the looseness of theoretical privacy bounds for DP-SGD, the de facto private training method in modern ML, as nearly-matching empirical lower bounds have been achieved under various threat models [NHSBTJCT23, AC24, CBP25]. In this work, we propose the empirical privacy lower bound of an algorithm as a concrete metric to optimize for, complementary to the theoretical upper bound. We give a lightweight defense framework that generically augments optimization methods in the ML pipeline to have significantly-improved empirical privacy on standard benchmarks. Moreover, we show that our framework comes at no theoretical privacy cost when augmenting DP-SGD, unlike previously-proposed defenses against membership inference attacks. We evaluate our defense against a broad range of audit constructions, models, and datasets to demonstrate its flexibility.
Arman Zareian Jahromi, Vishnu Bondalakunta, Mohammad Akbar Bin Shah +3cs.CR cs.LG
Image-to-image face generators are widely used, and visual dissimilarity between their outputs and source images is sometimes treated as evidence of privacy. Auditing whether these systems satisfy formal identity-level (epsilon, delta)-differential privacy requires choosing among several distinct routes for converting embedding-space observations into estimates or bounds on the differential privacy parameter epsilon. We present a comparative study of four such audits applicable to pre-trained, black-box face generators: a Gaussian-mechanism reading of per-identity sensitivity (GaussMech); a per-dimension kernel-density log-ratio aggregated by basic composition (KDE-LR); an analytical population-level lower bound on pure-DP epsilon derived from the maximum mean discrepancy via the total variation distance (MMD-TV); and a hypothesis-testing evaluation of a cross-validated classifier's out-of-fold ROC (ROC-HT). For each method we make explicit its assumptions, hyperparameter dependence, finite-sample limitations, and the regime in which its epsilon estimate is informative. Applied to FaceFusion and InstantID across multiple identity encoders and reference datasets, the audits consistently reveal substantial identity distinguishability while reporting markedly different epsilon estimates that reflect each method's distinct assumptions and finite-sample treatment. In this high-distinguishability regime, the experiments do not support a reliable ranking of the four methods. Their relative trade-offs should be evaluated on partially private mechanisms, which we identify as the natural next study. The resulting framework places these audits in a shared identity-level audit setting and clarifies how their assumptions and finite-sample treatments shape the resulting differential privacy estimates.
Black-box privacy scores for retrieval-augmented generation (RAG) are difficult to interpret unless the audited defense's active pipeline hook is known. We propose an active-path audit: inventory source-level hooks over retrieval, retrieved content, and generation; map each metric to the leakage channel it observes; and validate generated-text effects with exact-match canaries. In our benchmark reimplementations, the DP-style defenses modify retrieval scores only: their generation hooks are TODO-flagged stubs that return responses unchanged. This active path explains why they affect membership-inference behavior but track No-Defense on generated-text named-entity leakage, measured by NEL_strict. By contrast, the end-to-end LPRAG path is canary-validated on the email channel, recovering 53/150 canaries under No-Defense and 0/150 under LPRAG. These findings concern our reimplementations on our stack, not released defenses or defense families; the contribution is a methodology and case study, not a universal ranking
Offline Reinforcement Learning (RL) agents are trained on fixed behavioral trajectories, which makes trajectory-level deletion important when selected data must be removed after training. Evaluating such deletion is difficult because a lower membership score can reflect trajectory removal, residual memorization visible to another attack, or policy collapse that destroys useful behavior. We introduce Trajectory-level memOrization and Unlearning in offline RL (TOUR), a benchmark that combines trajectory-level partitioning, matched non-member controls, retraining references, retained-performance anchors, and multi-attack privacy auditing. Across D4RL locomotion experiments and an exploratory AntMaze extension, TOUR shows that common deletion baselines have environment-dependent privacy-utility behavior. Retraining and fine-tuning often provide stronger retained-utility references than uniform GA+Refit, while TrajDeleter remains a useful comparator but is not uniformly stronger under the same audit. Reference-model, threshold, deviation, equivalence, action-error, representation-based, and query-limited attacks further show that a single likelihood-based membership score can overstate deletion quality. In the evaluated settings, conclusions about offline RL unlearning are therefore not stable under single-score auditing. They depend on matched non-member construction, retraining-relative calibration, attack family, retained utility, and explicit scope for diagnostic architecture or component-level evidence.
Modern federated and streaming learning systems often release intermediate models, so privacy must hold for the full trajectory under adaptive interaction. Motivated by participation privacy, we study single-edit neighboring user streams, where one insertion/deletion shifts all subsequent updates and defeats standard Hamming-neighbor continual-release analyses. We give an auditable modular recipe. A randomized buffering wrapper emits bins of size $[U,2U]$, reducing single-edit streams to a Hamming-style per-bin update stream with explicit backlog/delay guarantees, where $U$ is calibrated by the privacy parameters $(\varepsilon,δ)$. We then prove a certification theorem identifying when a non-adaptive Hamming-neighbor DP proof for a continual primitive lifts to adaptive inputs: the primitive must use fresh per-round randomness and have a stable one-round privacy profile under common adaptive context. Together, these ingredients yield trajectory-level $(\varepsilon,δ)$-DP for single-edit streams using standard primitives (e.g., tree prefix sums), with an explicit privacy--latency link via $U$.
Evaluating whether unlearning algorithms truly remove training data influence remains an open challenge. We propose a practical auditor that computes data-dependent lower bounds on the unlearning parameter $\varepsilon$ using membership inference attacks. Evaluating multiple unlearning algorithms, we find a sharp separation: algorithms with rigorous guarantees, such as model clipping and rewind-to-delete, achieve very small $\varepsilon$ bounds that do not falsify their unlearning guarantees, whereas empirical methods such as Hessian-based unlearning, interleaved ascent-descent, ascent on the forget set, and fine-tuning on the retain set exhibit large bounds, indicating poor unlearning. Our auditor provides a practical tool for empirically falsifying unlearning claims through a hypothesis-testing framework, and we validate it on CIFAR-100 and Shakespeare text.
The tendency of large generative models to memorize training data makes sample verification critical for privacy auditing and copyright enforcement. Current membership (MIA) and dataset inference (DI) attacks often rely on one-shot generations, which yield weak signals and limited sensitivity across modalities. Inspired by Model Autophagy Disorder (MAD), we introduce MADreMIA, a model-agnostic framework that enhances white-, gray-, and black-box MIA and DI. Rather than relying on shadow model training -- often infeasible for large generative models -- our framework facilitates scalable inference by leveraging inherent signals through iterative trajectories. This process utilizes chained generations across diverse modalities, where each output serves as the subsequent input, to improve membership evidence at low FPR. We demonstrate that memorized training samples exhibit significantly higher coherence and slower degradation during iterative regeneration than non-member generations. Our results show that MADreMIA provides richer signals across diverse model families and modalities; we present comprehensive evaluations for IARs, diffusion, and language models, alongside preliminary results demonstrating its potential for audio models.
Lorenzo Rossi, Bartłomiej Marek, Franziska Boenisch +1cs.LG
Assessing the privacy of large language models (LLMs) presents significant challenges. In particular, most existing methods for auditing differential privacy require the insertion of specially crafted canary data during training, making them impractical for auditing already-trained models without costly retraining. Additionally, dataset inference, which audits whether a suspect dataset was used to train a model, is infeasible without access to a private non-member held-out dataset. Yet, such held-out datasets are often unavailable or difficult to construct for real-world cases since they have to be from the same distribution (IID) as the suspect data. These limitations severely hinder the ability to conduct scalable, post-hoc audits. To enable such audits, this work introduces natural identifiers (NIDs) as a novel solution to the above-mentioned challenges. NIDs are structured random strings, such as cryptographic hashes and shortened URLs, naturally occurring in common LLM training datasets. Their format enables the generation of unlimited additional random strings from the same distribution, which can act as alternative canaries for audits and as same-distribution held-out data for dataset inference. Our evaluation highlights that indeed, using NIDs, we can facilitate post-hoc differential privacy auditing without any retraining and enable dataset inference for any suspect dataset containing NIDs without the need for a private non-member held-out dataset.
The rapid adoption of generative AI and Large Language Models (LLMs) has spurred interest in synthetic data as a privacy-preserving alternative to sensitive real-world datasets. However, generating high-utility synthetic data often carries the risk of memorizing and regurgitating private information from the training corpus. In this work, we present a customizable empirical auditing framework designed to detect and explain such data disclosures. Our framework introduces a mechanism to distinguish between "true disclosures"-where the system directly reproduces a user's information-and "phantom disclosures''-where the system incidentally generates a user's data. By partitioning input data into training and holdout sets and applying rigorous statistical hypothesis testing, we determine if observed disclosures are consistent with strict privacy baselines, such as zero-learning or specific Differential Privacy (DP) bounds. Crucially, this approach requires no model access, no canary insertion, and no reference model training -only the synthetic output and a held-out control set. We demonstrate that this framework effectively functions as a membership inference attack, providing empirical lower bounds on privacy leakage that are tighter than prior data-based auditing methods. Our approach is model-agnostic, applies to any synthetic data generation mechanism, and requires orders of magnitude fewer computational resources than shadow-model or canary-based alternatives.
Privacy auditing provides an important safeguard by estimating the actual information leaked by a model, thus ensuring that theoretical privacy guarantees hold in practice. We study empirical privacy auditing for differentially private (DP) machine learning, focusing on efficient one-run methods for mechanisms such as DP-SGD. Prior one-run approaches threshold training examples or "canaries" into binary membership guesses, which discards useful information. We show that, in the white-box DP-SGD setting, canary-aligned signals naturally form a sequence of random variables whose normalized sum is asymptotically Gaussian. Leveraging this distributional perspective, we develop a DP-auditing framework that leads to tighter privacy lower bounds from a single training run.
Parameter-efficient fine-tuning of large language models (LLMs) can exhibit problematic memorization of individual training examples. Empirical privacy auditing (EPA) quantifies this risk by measuring realistic data leakage on membership inference (MI) or reconstruction attacks. A key challenge in EPA is designing ``canary'' examples that are mixed with the privacy-sensitive training data. We propose generating synthetic canaries via high-temperature sampling ($T \geq 0.8$) from LLMs, using prompts tailored to the privacy-sensitive training data. These canaries act as high-influence outliers, ensuring high identifiability and hence strong audits. Further, since the canaries are themselves non-private, they are inspectable and can be inserted with repetition without jeopardizing the privacy of the real data. An important use of models fine-tuned on privacy-sensitive data is the generation of synthetic data. This also comes with privacy risk. We introduce a powerful synthetic data audit based on fine-tuning an auxiliary model on the synthetic data. Auditing the auxiliary model for the original canaries then provides a strong estimate of the privacy leakage through the synthetic data. Finally, leveraging our strong auditing methodologies, we perform a systematic investigation into the interacting effects of model capacity and canary entropy on memorization.
Existing privacy analyses for Graph Neural Networks (GNNs) largely inherit assumptions from non-graph settings, overlooking structural correlations and stochastic training-graph sampling. In particular, node-dependent priors make type-I and type-II errors alone insufficient to characterize the best membership inference test. To address this, we introduce Bayesian Membership Privacy (BMP), a sampling-aware formulation of node-level membership privacy that incorporates node-dependent priors and treats graph sampling probabilities as part of the adversary's knowledge. BMP casts membership inference as a Bayesian hypothesis test and accordingly quantifies membership privacy in terms of posterior membership probability. We explore theoretical properties of BMP in relation to the existing definitions in the literature. We further propose a practical, sampling-aware auditing mechanism to estimate the parameters of BMP as a measure of node-level privacy leakage in GNNs. We conduct experiments on benchmark graph datasets and show that BMP yields fine-grained privacy insights that are not visible through global attack accuracy alone.