DAVSP: Safety Alignment for Large Vision-Language Models via Deep Aligned Visual Safety Prompt
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (44) · 38111-38119
DOI: 10.1609/aaai.v40i44.41149 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Yitong Zhang und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref).
Abstract (Originalsprache)
Large Vision-Language Models (LVLMs) have achieved impressive progress across various applications but remain vulnerable to malicious queries. Existing safety alignment approaches typically fail to resist malicious queries while preserving utility on benign ones effectively. To address these challenges, we propose DAVSP, which is built upon two key innovations. First, we introduce Visual Safety Prompt, which appends a trainable padding region around the input image. It preserves visual features and expands the optimization space. Second, we propose Deep Alignment, a novel approach to train the visual safety prompt through supervision in the model's activation space. It enhances the inherent ability of LVLMs to perceive malicious queries, achieving deeper alignment than prior works. Extensive experiments demonstrate that DAVSP effectively resists malicious queries while preserving benign input utility. Furthermore, DAVSP exhibits great cross-model generation ability. Ablation studies further reveal that both the Visual Safety Prompt and Deep Alignment are essential to the overall effectiveness.
Zitieren
APA
Zhang, Y., Li, J., Cai, L., & Li, G. (2026). DAVSP: Safety Alignment for Large Vision-Language Models via Deep Aligned Visual Safety Prompt. Proceedings of the AAAI Conference on Artificial Intelligence, 40(44), 38111-38119. https://doi.org/10.1609/aaai.v40i44.41149
Vancouver
Zhang Y, Li J, Cai L, Li G. DAVSP: Safety Alignment for Large Vision-Language Models via Deep Aligned Visual Safety Prompt. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(44):38111-38119. doi:10.1609/aaai.v40i44.41149
BibTeX
@article{Zhang2026,
author = {Zhang, Yitong and Li, Jia and Cai, Liyi and Li, Ge},
title = {DAVSP: Safety Alignment for Large Vision-Language Models via Deep Aligned Visual Safety Prompt},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {44},
pages = {38111--38119},
doi = {10.1609/aaai.v40i44.41149}
}
EUVIMED