Large vision-language models (LVLMs) derive their capabilities from extensive training on vast corpora of visual and textual data.

Deep learning models are vulnerable to privacy attacks due to their tendency to memorize individual training examples. Theoretically-sound defenses such as differential privacy can defend against this threat, but model performance often suffers. Empirical defenses may thwart existing attacks while maintaining model performance but do not offer any robust theoretical guarantees. In this paper, we explore a new strategy based on the concept of plausible deniability. We introduce a training algorithm called Plausibly Deniable Stochastic Gradient Descent (PD-SGD). The core of this approach is a rejection sampling technique, which probabilistically prevents updating model parameters whenever a mini-batch cannot be plausibly denied. We provide theoretical results showing that PD-SGD effectively mitigates privacy leakage from individual data points. Experiments demonstrate the scalability of PD-SGD and the favorable privacy-utility trade-off it offers compared to existing defense methods.

Large vision-language models (VLLMs) have driven significant progress in multimodal systems, enabling a wide range of applications across domains such as healthcare, education, and content generation. Despite the success, the large-scale datasets used to train these models often contain sensitive or personally identifiable information, raising serious privacy concerns. To audit and better understand such risks, membership inference attacks (MIAs) have become a key tool. However, existing MIAs against VLLMs predominantly assume access to full-model logits, which are typically unavailable in many practical deployments. To facilitate MIAs in a more realistic and restrictive setting, we propose a novel framework: label-only membership inference attacks (LOMIA) targeting pre-trained VLLMs where only the model's top-1 prediction is available. Within this framework, we propose three effective attack methods, all of which exploit the intuition that training samples are more likely to be memorized by the VLLMs, resulting in outputs that exhibit higher semantic alignment and lower perplexity. Our experiments show that our framework surpasses existing label-only attack adaptations for different VLLMs and competes with state-of-the-art logits-based attacks across all metrics on three widely used open-source VLLMs and GPT-4o.

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.

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.

Large vision-language models (VLLMs) have driven significant progress in multi-modal systems, enabling a wide range of applications across domains such as healthcare, education, and content generation. Despite the success, the large-scale datasets used to train these models often contain sensitive or personally identifiable information, raising serious privacy concerns. To audit and better understand such risks, membership inference attacks (MIAs) have become a key tool. However, existing MIAs against VLLMs predominantly assume access to full-model logits, which are typically unavailable in many practical deployments. To facilitate MIAs in a more realistic and restrictive setting, we propose a novel framework: label-only membership inference attacks (LOMIA) targeting pre-trained VLLMs where only the model's top-1 prediction is available. Within this framework, we propose three effective attack methods, all of which exploit the intuition that training samples are more likely to be memorized by the VLLMs, resulting in outputs that exhibit higher semantic alignment and lower perplexity. Our experiments show that our framework surpasses existing label-only attack adaptations for different VLLMs and competes with state-of-the-art logits-based attacks across all metrics on three widely used open-source VLLMs and GPT-4o.

We propose a novel and practical privacy notion called f-Membership Inference Privacy (f-MIP), which explicitly considers the capabilities of realistic adversaries under the membership inference attack threat model. Consequently, f-MIP offers interpretable privacy guarantees and improved utility (e.g., better classification accuracy). In particular, we derive a parametric family of f-MIP guarantees that we refer to as µ-Gaussian Membership Inference Privacy (µ-GMIP) by theoretically analyzing likelihood ratio-based membership inference attacks on stochastic gradient descent (SGD). Our analysis highlights that models trained with standard SGD already offer an elementary level of MIP. Additionally, we show how f-MIP can be amplified by adding noise to gradient updates.

Membership inference attacks are designed to determine, using black box access to trained models, whether a particular example was used in training or not. Membership inference can be formalized as a hypothesis testing problem. The most effective existing attacks estimate the distribution of some test statistic (usually the model's confidence on the true label) on points that were (and were not) used in training by training many shadow models--i.e.

Membership Inference Attacks (MIA) aim to infer whether a target data record has been utilized for model training or not. Existing MIAs designed for large language models (LLMs) can be bifurcated into two types: reference-free and reference-based attacks. Although reference-based attacks appear promising performance by calibrating the probability measured on the target model with reference models, this illusion of privacy risk heavily depends on a reference dataset that closely resembles the training set. Both two types of attacks are predicated on the hypothesis that training records consistently maintain a higher probability of being sampled. However, this hypothesis heavily relies on the overfitting of target models, which will be mitigated by multiple regularization methods and the generalization of LLMs. Thus, these reasons lead to high false-positive rates of MIAs in practical scenarios.We propose a Membership Inference Attack based on Self-calibrated Probabilistic Variation (SPV-MIA). Specifically, we introduce a self-prompt approach, which constructs the dataset to fine-tune the reference model by prompting the target LLM itself. In this manner, the adversary can collect a dataset with a similar distribution from public APIs.Furthermore, we introduce probabilistic variation, a more reliable membership signal based on LLM memorization rather than overfitting, from which we rediscover the neighbour attack with theoretical grounding. Comprehensive evaluation conducted on three datasets and four exemplary LLMs shows that SPV-MIA raises the AUC of MIAs from 0.7 to a significantly high level of 0.9.