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

arXiv: https://arxiv.org/abs/2508.18649 · HTML: https://mlsanity.com/paper/2508.18649/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 18 of its 36 references that are in the map.
- First posted on arXiv 26 Aug 2025.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)
- Multimodal LLMs / vision-language models / [MLLM jailbreaks and adversarial robustness](https://mlsanity.com/mllm/mllm-jailbreaks-and-adversarial-robustness/)
- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)

## Builds on (most cited first, 18 of 18)

- [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution](https://mlsanity.com/paper/2409.12191/) (2024)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models](https://mlsanity.com/paper/2504.10479/) (2025)
- [Self-Rewarding Language Models](https://mlsanity.com/paper/2401.10020/) (2024)
- [Deliberative Alignment: Reasoning Enables Safer Language Models](https://mlsanity.com/paper/2412.16339/) (2024)
- [JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks](https://mlsanity.com/paper/2404.03027/) (2024)
- [MuirBench: A Comprehensive Benchmark for Robust Multi-image Understanding](https://mlsanity.com/paper/2406.09411/) (2024)
- [Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models](https://mlsanity.com/paper/2402.02207/) (2024)
- [STAIR: Improving Safety Alignment with Introspective Reasoning](https://mlsanity.com/paper/2502.02384/) (2025)
- [STAR-1: Safer Alignment of Reasoning LLMs with 1K Data](https://mlsanity.com/paper/2504.01903/) (2025)
- [ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time](https://mlsanity.com/paper/2410.06625/) (2024)
- [Multimodal Situational Safety](https://mlsanity.com/paper/2410.06172/) (2024)
- [MM-Vet v2: A Challenging Benchmark to Evaluate Large Multimodal Models for Integrated Capabilities](https://mlsanity.com/paper/2408.00765/) (2024)
- [IDEATOR: Jailbreaking and Benchmarking Large Vision-Language Models Using Themselves](https://mlsanity.com/paper/2411.00827/) (2024)
- [Attacking Adversarial Attacks as A Defense](https://mlsanity.com/paper/2106.04938/) (2021)
- [Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models](https://mlsanity.com/paper/2501.18533/) (2025)
- [FC-Attack: Jailbreaking Multimodal Large Language Models via Auto-Generated Flowcharts](https://mlsanity.com/paper/2502.21059/) (2025)
- [PuriDefense: Randomized Local Implicit Adversarial Purification for Defending Black-box Query-based Attacks](https://mlsanity.com/paper/2401.10586/) (2024)

## Built on by (most cited first, 2 of 2)

- [Towards Safe Reasoning in Large Reasoning Models via Corrective Intervention](https://mlsanity.com/paper/2509.24393/) (2025)
- [LPG: Balancing Efficiency and Policy Reasoning in Latent Policy Guardrails](https://mlsanity.com/paper/2605.17329/) (2026)

## Abstract (arXiv)

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.