[2603.28378] Membership Inference Attacks against Large Audio Language Models

Computer Science > Sound arXiv:2603.28378 (cs) [Submitted on 30 Mar 2026] Title:Membership Inference Attacks against Large Audio Language Models Authors:Jia-Kai Dong, Yu-Xiang Lin, Hung-Yi Lee View a PDF of the paper titled Membership Inference Attacks against Large Audio Language Models, by Jia-Kai Dong and 2 other authors View PDF HTML (experimental) Abstract:We present the first systematic Membership Inference Attack (MIA) evaluation of Large Audio Language Models (LALMs). As audio encodes non-semantic information, it induces severe train and test distribution shifts and can lead to spurious MIA performance. Using a multi-modal blind baseline based on textual, spectral, and prosodic features, we demonstrate that common speech datasets exhibit near-perfect train/test separability (AUC approximately 1.0) even without model inference, and the standard MIA scores strongly correlate with these blind acoustic artifacts (correlation greater than 0.7). Using this blind baseline, we identify that distribution-matched datasets enable reliable MIA evaluation without distribution shift confounds. We benchmark multiple MIA methods and conduct modality disentanglement experiments on these datasets. The results reveal that LALM memorization is cross-modal, arising only from binding a speaker's vocal identity with its text. These findings establish a principled standard for auditing LALMs beyond spurious correlations. Comments: Subjects: Sound (cs.SD); Artificial Intelligence (cs.AI) A...