arXiv Machine Learning

Seeing Through Conflicts: Improving Instruction Hierarchy Alignment in Vision-Language Models

arXiv AI
Sep 7

MM-IFEval-Pro: A Multilingual and Attack-Resistant Benchmark for Instruction-Following in Vision-Language Models

MM-IFEval-Pro is a new multilingual benchmark for evaluating instruction-following in vision-language models, covering both Chinese and English tasks. It includes 4 major task categories, 24 subcategories, and 8 instruction categories with 52 subcategories, each sample featuring an average of 3.0 constraints to mimic complex instruction scenarios. A reinforcement-learning training set with Chinese and adversarial instructions improves model performance on MM-IFEval-Pro and transfers well to other multimodal benchmarks, showing strong cross-task and cross-language generalization.

By Changming Xiao, Zhenliang Ni, Jinhui He, Han Shu, Jie Hu
arXiv AI
Aug 12

SafeCap: Improving LVLM Safety with Image Captioning Reinforcement Learning

arXiv:2608. 10513v1 Announce Type: cross Abstract: Large vision-language models (LVLMs) remain vulnerable to jailbreak attacks that exploit visual inputs to bypass safety alignment inherited from their language backbones.

By Caoyuan Ma, Wenpu Liu, Weichu Xie, Tian Gu, Shilei Zhao, Lingxi Min, Shuai Dong, Yuqi Xu, Ji Zhao, Ziyue Wang, Wenzheng Chang, Taiqiang Wu, Yongfu Zhu, Wenqi Shao, Yinqiang Zheng
Hugging Face Trending Papers
Aug 19

Breaking the weakest link to evade vision language models

The paper investigates how Vision Language Models (VLMs) can be fooled by tiny, human‑imperceptible changes to images. It introduces a gradient‑based attack that targets only the vision encoder, reducing computational cost while still effectively disrupting both untargeted and targeted multimodal interpretations. Experiments on open‑source VLMs such as Qwen2.5‑VL, Granite‑Vision, FastVLM, and Phi‑3.5‑Vision demonstrate that these small perturbations can dramatically alter the models’ textual outputs.

Hugging Face Trending Papers
Aug 19

When Safety Overrides Vision: Exploring Dynamics between Vision Influence and Safety Alignment in Vision-Language Models

Aligned vision‑language models (VLMs) are designed to combine grounded visual reasoning with safe generation. The study finds that when safety constraints are applied, these models often abstain from answering questions that they could answer under default instruction, yet visual evidence continues to influence the decoding process. The authors show that safety‑induced abstention alters late‑stage hidden‑state dynamics, and that targeted interventions can restore grounded answering without retraining or changing visual inputs.

arXiv AI
Aug 24

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

ReFrame is a training‑free framework that enhances safety alignment for multimodal large language models at test time. It uses two lightweight agents: one generates risk and utility evidence, and the other rewrites prompts and routes images to create a safe proxy before invoking the deployed MLLM. Experiments show that ReFrame improves jailbreak defense, safety awareness, and reduces over‑sensitivity while maintaining multimodal utility.

By Wenzheng Jiang, Xuankun Rong, Yuanzhao Zhai, Dawei Feng, Huaimin Wang
arXiv AI
Aug 20

Breaking the weakest link to evade vision language models

The paper investigates how Vision Language Models (VLMs) can be fooled by small, human‑imperceptible changes to images. It introduces a gradient‑based attack that targets only the vision encoder, reducing computational cost while still effectively disrupting both untargeted and targeted multimodal alignment. Experiments on open‑source VLMs such as Qwen2.5‑VL, Granite‑Vision, FastVLM, and Phi‑3.5‑Vision demonstrate that these perturbations can significantly alter the models’ textual outputs.

By Ilan Zini, Boussad Addad, Katarzyna Kapusta