EUVIMEDEUVIMED entdecken
Crossref

Rethinking Membership Inference Attacks for CLIP

Lluis Gomez

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (25) · 21305-21313

DOI: 10.1609/aaai.v40i25.39276 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Lluis Gomez.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Membership Inference Attacks (MIAs) test whether a model has memorized training data, and are a key tool for auditing privacy risks in machine learning. Recent papers report near-perfect MIA success against large vision-language models such as CLIP, but almost all evaluations train on one web-scale corpus (e.g. LAION-400M) and treat samples from a different corpus (e.g. COCO or CC12M) as non-members - thereby turning the task into out-of-distribution (OOD) detection rather than true membership testing, introducing spurious signals unrelated to true memorization. We revisit the problem with a distribution-matched benchmark built from the CommonPool-L corpus of DataComp. A ViT-B/16 CLIP trained on 400M pairs is accompanied by two 26-shard, i.i.d. splits that serve as member and non-member sets, sharing the exact same acquisition and preprocessing pipeline. Under this strictly in-distribution setting, every published MIA baseline collapses to chance (~51% AUC). To explain this collapse, we derive a scaling-law upper bound for similarity-based attacks showing that the expected member vs. non-member similarity gap decays as O(T/N) for contrastive learning with T epochs over N samples. Empirically, as we vary the training set size while holding all hyper-parameters fixed, the gap follows the predicted linear trend in log–log space, and Cosine Similarity Attack AUC drops from 94% to 51%. Finally, we propose a simple, white-box, gradient-based MIA that outperforms prior attacks for CLIP without relying on OOD cues. We release code, checkpoints, and data to foster comprehensive and reproducible privacy research on multimodal CLIP-like foundation models.

Zitieren

APA

Gomez, L. (2026). Rethinking Membership Inference Attacks for CLIP. Proceedings of the AAAI Conference on Artificial Intelligence, 40(25), 21305-21313. https://doi.org/10.1609/aaai.v40i25.39276

Vancouver

Gomez L. Rethinking Membership Inference Attacks for CLIP. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(25):21305-21313. doi:10.1609/aaai.v40i25.39276

BibTeX
@article{Gomez2026,
  author = {Gomez, Lluis},
  title = {Rethinking Membership Inference Attacks for CLIP},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {25},
  pages = {21305--21313},
  doi = {10.1609/aaai.v40i25.39276}
}
Kostenlos starten →