analiza aiqbench
This research establishes a critical auditing standard for Large Audio Language Models by proving that membership inference attacks often detect dataset distribution shifts rather than actual model memorization. The study reveals that cross-modal binding of vocal identity with text is the primary source of vulnerability, offering a principled protocol to distinguish genuine privacy risks from spurious correlations in audio AI systems.
Membership Inference Attacks against Large Audio Language Models
arXiv:2603.28378v2 Announce Type: replace-cross Abstract: We present the first systematic Membership Inference Attack (MIA) evaluation of LALMs. Using Multi-modal Blind Baselines based on textual, spectral and prosodic features, we demonstrate that common audio datasets exhibit near-perfect train/test separability (AUC ~ 1.0) even without model inference, thus MIA may primarily detect distribution shift. We therefore introduce a blind-baseline protocol to control for this confound. Under this protocol, we identify that the distribution-matched datasets enable reliable MIA evaluation without distribution-shift artifacts. 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. Our codebase is available at https://github.com/snooow1029/ALM_MIA.
Abstract:We present the first systematic Membership Inference Attack (MIA) evaluation of LALMs. Using Multi-modal Blind Baselines based on textual, spectral and prosodic features, we demonstrate that common audio datasets exhibit near-perfect train/test separability (AUC ~ 1.0) even without model inference, thus MIA may primarily detect distribution shift. We therefore introduce a blind-baseline protocol to control for this confound. Under this protocol, we identify that the distribution-matched datasets enable reliable MIA evaluation without distribution-shift artifacts. 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. Our codebase is available at this https URL.
Sound (cs.SD); Artificial Intelligence (cs.AI)
I.2.6; I.5.2; K.4.1
arXiv:2603.28378 [cs.SD]
(or arXiv:2603.28378v2 [cs.SD] for this version)
https://doi.org/10.48550/arXiv.2603.28378 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 223 cuvinte
Subiecte conexe
Impact of Graph Structure on Membership-Inference Risk for Graph Neural Networks
arXiv · 3 iun. 2026
From Representation to Readout: Mechanistic Analysis of Membership Signals in Language Models
arXiv · 2 nov. 2026
The Sample Complexity of Membership Inference and Privacy Auditing
arXiv · 9 oct. 2026
A Simple Presentation of Robust Membership Inference Attacks
arXiv · 28 oct. 2026