EUVIMEDEUVIMED entdecken
Crossref

Unveiling the Attribute Misbinding Threat in Identity-Preserving Models

Junming Fu, Jishen Zeng, Yi Jiang, Peiyu Zhuang, Baoying Chen, Siyu Lu, Jianquan Yang

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (1) · 283-291

DOI: 10.1609/aaai.v40i1.36989 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Junming Fu und 6 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Identity-preserving models have led to notable progress in generating personalized content. Unfortunately, such models also exacerbate risks when misused, for instance, by generating threatening content targeting specific individuals. This paper introduces the Attribute Misbinding Attack, a novel method that poses a threat to identity-preserving models by inducing them to produce Not-Safe-For-Work (NSFW) content. The attack's core idea involves crafting benign-looking textual prompts to circumvent text-filter safeguards and leverage a key model vulnerability: flawed attribute binding that stems from its internal attention bias. This results in misattributing harmful descriptions to a target identity and generating NSFW outputs. To facilitate the study of this attack, we present the Misbinding Prompt evaluation set, which examines the content generation risks of current state-of-the-art identity-preserving models across four risk dimensions: pornography, violence, discrimination, and illegality. Additionally, we introduce the Attribute Binding Safety Score (ABSS), a metric for concurrently assessing both content fidelity and safety compliance. Experimental results show that our Misbinding Prompt evaluation set achieves a 5.28 % higher success rate in bypassing five leading text filters (including GPT-4o) compared to existing main-stream evaluation sets, while also demonstrating the highest proportion of NSFW content generation. The proposed ABSS metric enables a more comprehensive evaluation of identity-preserving models by concurrently assessing both content fidelity and safety compliance.

Zitieren

APA

Fu, J., Zeng, J., Jiang, Y., Zhuang, P., Chen, B., Lu, S., & Yang, J. (2026). Unveiling the Attribute Misbinding Threat in Identity-Preserving Models. Proceedings of the AAAI Conference on Artificial Intelligence, 40(1), 283-291. https://doi.org/10.1609/aaai.v40i1.36989

Vancouver

Fu J, Zeng J, Jiang Y, Zhuang P, Chen B, Lu S, et al. Unveiling the Attribute Misbinding Threat in Identity-Preserving Models. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(1):283-291. doi:10.1609/aaai.v40i1.36989

BibTeX
@article{Fu2026,
  author = {Fu, Junming and Zeng, Jishen and Jiang, Yi and Zhuang, Peiyu and Chen, Baoying and Lu, Siyu and Yang, Jianquan},
  title = {Unveiling the Attribute Misbinding Threat in Identity-Preserving Models},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {1},
  pages = {283--291},
  doi = {10.1609/aaai.v40i1.36989}
}
Kostenlos starten →