State-of-the-art membership inference attacks (MIAs) typically require training many reference models, making it difficult to scale these attacks to large pre-trained language models (LLMs). As a result, prior research has either relied on weaker attacks that avoid training references (e.g., fine-tuning attacks), or on stronger attacks applied to small models and datasets. However, weaker attacks have been shown to be brittle and insights from strong attacks in simplified settings do not translate to today's LLMs. These challenges prompt an important question: are the limitations observed in prior work due to attack design choices, or are MIAs fundamentally ineffective on LLMs? We address this question by scaling LiRA--one of the strongest MIAs--to GPT-2 architectures ranging from 10M to 1B parameters, training references on over 20B tokens from the C4 dataset. Our results advance the understanding of MIAs on LLMs in four key ways. While (1) strong MIAs can succeed on pretrained LLMs, (2) their effectiveness, remains limited (e.g., AUC<0.7) in practical settings.

Recent works have presented methods to identify if individual text sequences were members of the model's training data, known as membership inference attacks (MIAs). We demonstrate that the apparent success of these MIAs is confounded by selecting non-members (text sequences not used for training) belonging to a different distribution from the members (e.g., temporally shifted recent Wikipedia articles compared with ones used to train the model). This distribution shift makes membership inference appear successful. However, most MIA methods perform no better than random guessing when discriminating between members and non-members from the same distribution (e.g., in this case, the same period of time).Even when MIAs work, we find that different MIAs succeed at inferring membership of samples from different distributions.Instead, we propose a new dataset inference method to accurately identify the datasets used to train large language models.

Membership Inference Attacks (MIAs) are widely used to quantify training data memorization and assess privacy risks. Standard evaluation requires repeated retraining, which is computationally costly for large models. One-run methods (single training with randomized data inclusion) and zero-run methods (post hoc evaluation) are often used instead, though their statistical validity remains unclear. To address this gap, we frame MIA evaluation as a causal inference problem, defining memorization as the causal effect of including a data point in the training set. This novel formulation reveals and formalizes key sources of bias in existing protocols: one-run methods suffer from interference between jointly included points, while zero-run evaluations popular for LLMs are confounded by non-random membership assignment. We derive causal analogues of standard MIA metrics and propose practical estimators for multi-run, one-run, and zero-run regimes with non-asymptotic consistency guarantees. Experiments on real-world data show that our approach enables reliable memorization measurement even when retraining is impractical and under distribution shift, providing a principled foundation for privacy evaluation in modern AI systems.

Generative audio models, based on diffusion and autoregressive architectures, have advanced rapidly in both quality and expressiveness. This progress, however, raises pressing copyright concerns, as such models are often trained on vast corpora of artistic and commercial works. A central question is whether one can reliably verify if an artist's material was included in training, thereby providing a means for copyright holders to protect their content. In this work, we investigate the feasibility of such verification through membership inference attacks (MIA) on open-source generative audio models, which attempt to determine whether a specific audio sample was part of the training set. Our empirical results show that membership inference alone is of limited effectiveness at scale, as the per-sample membership signal is weak for models trained on large and diverse datasets. However, artists and media owners typically hold collections of works rather than isolated samples. Building on prior work in text and vision domains, in this work we focus on dataset inference (DI), which aggregates diverse membership evidence across multiple samples. We find that DI is successful in the audio domain, offering a more practical mechanism for assessing whether an artist's works contributed to model training. Our results suggest DI as a promising direction for copyright protection and dataset accountability in the era of large audio generative models.

Multi-domain graph pre-training has emerged as a pivotal technique in developing graph foundation models. While it greatly improves the generalization of graph neural networks, its privacy risks under membership inference attacks (MIAs), which aim to identify whether a specific instance was used in training (member), remain largely unexplored. However, effectively conducting MIAs against multi-domain graph pre-trained models is a significant challenge due to: (i) Enhanced Generalization Capability: Multi-domain pre-training reduces the overfitting characteristics commonly exploited by MIAs. (ii) Unrepresentative Shadow Datasets: Diverse training graphs hinder the obtaining of reliable shadow graphs. (iii) Weakened Membership Signals: Embedding-based outputs offer less informative cues than logits for MIAs. To tackle these challenges, we propose MGP-MIA, a novel framework for Membership Inference Attacks against Multi-domain Graph Pre-trained models. Specifically, we first propose a membership signal amplification mechanism that amplifies the overfitting characteristics of target models via machine unlearning. We then design an incremental shadow model construction mechanism that builds a reliable shadow model with limited shadow graphs via incremental learning. Finally, we introduce a similarity-based inference mechanism that identifies members based on their similarity to positive and negative samples. Extensive experiments demonstrate the effectiveness of our proposed MGP-MIA and reveal the privacy risks of multi-domain graph pre-training.

Membership inference attacks (MIAs) test whether a data point was part of a model's training set, posing serious privacy risks. Existing methods often depend on shadow models or heavy query access, which limits their practicality. We propose GP-MIA, an efficient and interpretable approach based on Gaussian process (GP) meta-modeling. Using post-hoc metrics such as accuracy, entropy, dataset statistics, and optional sensitivity features (e.g. gradients, NTK measures) from a single trained model, GP-MIA trains a GP classifier to distinguish members from non-members while providing calibrated uncertainty estimates. Experiments on synthetic data, real-world fraud detection data, CIFAR-10, and WikiText-2 show that GP-MIA achieves high accuracy and generalizability, offering a practical alternative to existing MIAs.

Membership inference attacks (MIAs) have emerged as the standard tool for evaluating the privacy risks of AI models. However, state-of-the-art attacks require training numerous, often computationally expensive, reference models, limiting their practicality. We present a novel approach for estimating model-level vulnerability, the TPR at low FPR, to membership inference attacks without requiring reference models. Empirical analysis shows loss distributions to be asymmetric and heavy-tailed and suggests that most points at risk from MIAs have moved from the tail (high-loss region) to the head (low-loss region) of the distribution after training. We leverage this insight to propose a method to estimate model-level vulnerability from the training and testing distribution alone: using the absence of outliers from the high-loss region as a predictor of the risk. We evaluate our method, the TNR of a simple loss attack, across a wide range of architectures and datasets and show it to accurately estimate model-level vulnerability to the SOTA MIA attack (LiRA). We also show our method to outperform both low-cost (few reference models) attacks such as RMIA and other measures of distribution difference. We finally evaluate the use of non-linear functions to evaluate risk and show the approach to be promising to evaluate the risk in large-language models.

Abstract--As large language models (LLMs) become progressively more embedded in clinical decision-support, documentation, and patient-information systems, ensuring their privacy and trustworthiness has emerged as an imperative challenge for the healthcare sector . Fine-tuning LLMs on sensitive electronic health record (EHR) data improves domain alignment but also raises the risk of exposing patient information through model behaviors. In this work-in-progress, we present an exploratory empirical study on membership inference vulnerabilities in clinical LLMs, focusing on whether adversaries can infer if specific patient records were used during model training. Using a state-of-the-art clinical question-answering model, Llemr, we evaluate both canonical loss-based attacks and a domain-motivated paraphrasing-based perturbation strategy that more realistically reflects clinical adversarial conditions. Our preliminary findings reveal limited but measurable membership leakage, suggesting that current clinical LLMs provide partial resistance yet remain susceptible to subtle privacy risks that could undermine trust in clinical AI adoption. These results motivate continued development of context-aware, domain-specific privacy evaluations and defenses such as differential privacy fine-tuning and paraphrase-aware training, to strengthen the security and trustworthiness of healthcare AI systems. Large language models (LLMs) have become essential components across a broad spectrum of modern computational systems [1].