ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models

Abstract

While multimodal large language models (MLLMs) extend model capabilities beyond text, they also make safety alignment increasingly challenging. Multimodal safety alignment methods must address cross-modal jailbreaks, safety-awareness failures, and over-sensitive refusals. However, existing methods often rely on retraining or internal-state inspection, limiting their applicability to deployed closed-source MLLMs and motivating test-time safety alignment. We analyze this setting and identify two k

Categories

Framework mappings

Suggested from the entry's categories.

@misc{jiang2026reframe,
  title = {{ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models}},
  author = {Wenzheng Jiang and Xuan-Kun Rong and Yuan-Zhao Zhai and Da-Wei Feng and Huaimin Wang},
  year = {2026},
  month = aug,
  eprint = {2608.21100},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/8216bd75e51ed23466c2d4f47ea1dfe2f32e75fd}
}