PRISM: Robust VLM Alignment with Principled Reasoning for Integrated Safety in Multimodality

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Nanxi, Zhao, Zhengyue, Suh, G. Edward, Pavone, Marco, Xiao, Chaowei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911561324429312
author Li, Nanxi
Zhao, Zhengyue
Suh, G. Edward
Pavone, Marco
Xiao, Chaowei
author_facet Li, Nanxi
Zhao, Zhengyue
Suh, G. Edward
Pavone, Marco
Xiao, Chaowei
contents Safeguarding vision-language models (VLMs) is a critical challenge, as existing methods often suffer from over-defense, which harms utility, or rely on shallow alignment, failing to detect complex threats that require deep reasoning. To this end, we introduc PRISM (Principled Reasoning for Integrated Safety in Multimodality), a System 2-like framework that aligns VLMs through a structured four-stage reasoning process explicitly designed to handle three distinct categories of multimodal safety violations. Our framework consists of two key components: a structured reasoning pipeline that analyzes each violation category in dedicated stages, and PRISM-DPO, generated via Monte Carlo Tree Search (MCTS) to refine reasoning quality through Direct Preference Optimization. Comprehensive evaluations show that PRISM substantially reduces attack success rates on JailbreakV-28K and VLBreak, improves robustness against adaptive attacks, and generalizes to out-of-distribution multi-image threats, while better preserving model utility on benign multimodal benchmarks. Our code, data, and model weights available at https://github.com/SaFoLab-WISC/PRISM.
format Preprint
id arxiv_https___arxiv_org_abs_2508_18649
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PRISM: Robust VLM Alignment with Principled Reasoning for Integrated Safety in Multimodality
Li, Nanxi
Zhao, Zhengyue
Suh, G. Edward
Pavone, Marco
Xiao, Chaowei
Cryptography and Security
Artificial Intelligence
Safeguarding vision-language models (VLMs) is a critical challenge, as existing methods often suffer from over-defense, which harms utility, or rely on shallow alignment, failing to detect complex threats that require deep reasoning. To this end, we introduc PRISM (Principled Reasoning for Integrated Safety in Multimodality), a System 2-like framework that aligns VLMs through a structured four-stage reasoning process explicitly designed to handle three distinct categories of multimodal safety violations. Our framework consists of two key components: a structured reasoning pipeline that analyzes each violation category in dedicated stages, and PRISM-DPO, generated via Monte Carlo Tree Search (MCTS) to refine reasoning quality through Direct Preference Optimization. Comprehensive evaluations show that PRISM substantially reduces attack success rates on JailbreakV-28K and VLBreak, improves robustness against adaptive attacks, and generalizes to out-of-distribution multi-image threats, while better preserving model utility on benign multimodal benchmarks. Our code, data, and model weights available at https://github.com/SaFoLab-WISC/PRISM.
title PRISM: Robust VLM Alignment with Principled Reasoning for Integrated Safety in Multimodality
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2508.18649