TabPATE: Differentially Private Tabular In-Context Learning Without Public Data

Tabular foundation models are commonly assumed to present limited privacy concerns as they are often pre-trained on large collections of synthetic data. However, these models leverage in-context learning, where sensitive records may be provided directly at inference time as labelled context examples. In this paper, we demonstrate that predictions generated via the attention mechanism leak sufficient information to enable effective Membership Inference Attacks (MIAs). To highlight this vulnerability, we propose AMIA (Attention-based Membership Inference Attack), a shadow-model-free attack that exploits the concentration of transformer attention patterns. Our results show that attention mechanisms reveal strong membership signals, which exceed classical confidence-based attacks, achieving an average gain of 7.7%, specially in low false-positive regimes. To mitigate this risk, we introduce an inference-time defence inspired by $k$-anonymity principles. This approach reduces the uniqueness of context-key representations without introducing random noise or retraining the model. By targeting only high-risk queries identified through AMIA scores, the defence substantially reduces membership leakage of this attack by an average of 50% and 25% against confidence-based attacks, while preserving predictive utility with only 3.9% performance degradation. Beyond showing that context examples are vulnerable, we further demonstrate that fine-tuning introduces an additional source of privacy risk. In particular, samples whose prediction confidence increases after fine-tuning become more susceptible to MIAs, indicating that fine-tuning can amplify memorisation and expose sensitive training information through confidence shifts.

Tabular data plays an important role in many fields and industries, including those with elevated privacy considerations and risks. As such, there is a rising interest in generating high-quality synthetic proxies for real tabular data as a means of reducing privacy risk and proprietary data exposure. With tabular diffusion models (TDMs) demonstrating leading performance in synthesizing such data, understanding and measuring the privacy risks associated with these models is imperative. Leveraging state-of-the-art membership inference attacks for TDMs in both black- and white-box settings, this work quantifies the impact of training setup, synthesis choices, and attacker knowledge on privacy leakage. Moreover, the results demonstrate that adversaries need not have perfect knowledge of the training setup, identical data distributions, or massive compute resources to construct successful attacks. Finally, the pitfalls associated with applying heuristic privacy metrics, such as distance-to-closest record, are revealed.

Generating high-fidelity synthetic tabular data under formal differential privacy guarantees remains an open challenge. Methods that provide strong theoretical protection typically sacrifice the modeling of inter-feature dependencies required for realistic synthesis, while architectures that excel at capturing complex column relationships offer only empirical privacy guarantees. We present PATE-TabTransGAN, a generative framework that integrates the Private Aggregation of Teacher Ensembles (PATE) mechanism with a Transformer-based student discriminator to jointly address both requirements, and employs a GNMax RDP accountant for numerically stable privacy accounting. An ensemble of Logistic Regression teachers trained on disjoint partitions supervise the student via noisy-aggregated labels, and a residual generator is optimized against this differentially private student, inheriting formal (ε, δ)-DP guarantees by post-processing. PATE-TabTransGAN was compared with PATE-GAN, DP-GAN, and DP-CTGAN, considered state-of-the-art in differentially private tabular synthesis. Experiments conducted on four tabular benchmarks (Adult, Breast, Cardio, Cervical) confirmed the high quality of the proposed method: PATE-TabTransGAN attains the best or tied-best AUROC on all four datasets. On AUCPR it matches the strongest baseline on Cardio, leads on Cervical, and trails on Breast; on Adult, we demonstrate that AUCPR is highly sensitive to positive-class convention, and that the observed gap is consistent with a convention difference between evaluation pipelines rather than a synthesis deficit.