VidLeaks: Membership Inference Attacks Against Text-to-Video Models

2026-01-16

AI Summary

This work addresses the privacy and copyright risks posed by potential data leakage in text-to-video (T2V) models, a challenge exacerbated by the spatiotemporal complexity of video generation that renders existing membership inference attacks ineffective. To this end, we propose VidLeaks, the first systematic framework for membership inference against T2V models. VidLeaks leverages two complementary signals—Spatial Reconstruction Fidelity (SRF) and Temporal Generation Stability (TGS)—to uncover how models memorize sparse keyframes and stochastic dynamics. Our approach enhances spatial signals via Top-K similarity and captures temporal leakage through multi-round query semantic consistency, operating effectively under supervised, reference-based, and query-only black-box settings. Experiments on AnimateDiff and InstructVideo demonstrate strong attack performance, achieving AUC scores of 82.92% and 97.01%, respectively, even in the strict query-only setting, thereby revealing significant privacy vulnerabilities.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The proliferation of powerful Text-to-Video (T2V) models, trained on massive web-scale datasets, raises urgent concerns about copyright and privacy violations. Membership inference attacks (MIAs) provide a principled tool for auditing such risks, yet existing techniques - designed for static data like images or text - fail to capture the spatio-temporal complexities of video generation. In particular, they overlook the sparsity of memorization signals in keyframes and the instability introduced by stochastic temporal dynamics. In this paper, we conduct the first systematic study of MIAs against T2V models and introduce a novel framework VidLeaks, which probes sparse-temporal memorization through two complementary signals: 1) Spatial Reconstruction Fidelity (SRF), using a Top-K similarity to amplify spatial memorization signals from sparsely memorized keyframes, and 2) Temporal Generative Stability (TGS), which measures semantic consistency across multiple queries to capture temporal leakage. We evaluate VidLeaks under three progressively restrictive black-box settings - supervised, reference-based, and query-only. Experiments on three representative T2V models reveal severe vulnerabilities: VidLeaks achieves AUC of 82.92% on AnimateDiff and 97.01% on InstructVideo even in the strict query-only setting, posing a realistic and exploitable privacy risk. Our work provides the first concrete evidence that T2V models leak substantial membership information through both sparse and temporal memorization, establishing a foundation for auditing video generation systems and motivating the development of new defenses. Code is available at: https://zenodo.org/records/17972831.

Problem

• Membership Inference Attacks
• Text-to-Video Models
• Privacy Leakage
• Spatio-temporal Memorization
• Video Generation

Innovation

• Membership Inference Attack
• Text-to-Video Models
• Spatial Reconstruction Fidelity
• Temporal Generative Stability
• Sparse-Temporal Memorization