# Attention Misses Visual Risk: Risk-Adaptive Steering for Multimodal Safety Alignment

arXiv: https://arxiv.org/abs/2510.13698 · HTML: https://mlsanity.com/paper/2510.13698/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 45 of its 57 references that are in the map.
- First posted on arXiv 15 Oct 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)
- LLM safety and security / [Vision-language and text-to-image safety](https://mlsanity.com/safety/vision-language-and-text-to-image-safety/)

## Builds on (most cited first, 30 of 45)

- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [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)
- [CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features](https://mlsanity.com/paper/1905.04899/) (2019)
- [LLaVA-OneVision: Easy Visual Task Transfer](https://mlsanity.com/paper/2408.03326/) (2024)
- [InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks](https://mlsanity.com/paper/2312.14238/) (2023)
- [Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond](https://mlsanity.com/paper/2308.12966/) (2023)
- [Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering](https://mlsanity.com/paper/2209.09513/) (2022)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilities](https://mlsanity.com/paper/2308.02490/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models](https://mlsanity.com/paper/2308.01263/) (2023)
- [Safety Alignment Should Be Made More Than Just a Few Tokens Deep](https://mlsanity.com/paper/2406.05946/) (2024)
- [Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation](https://mlsanity.com/paper/2310.06987/) (2023)
- [FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts](https://mlsanity.com/paper/2311.05608/) (2023)
- [Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions](https://mlsanity.com/paper/2309.07875/) (2023)
- [Visual Adversarial Examples Jailbreak Aligned Large Language Models](https://mlsanity.com/paper/2306.13213/) (2023)
- [MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models](https://mlsanity.com/paper/2311.17600/) (2023)
- [In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering](https://mlsanity.com/paper/2311.06668/) (2023)
- [OR-Bench: An Over-Refusal Benchmark for Large Language Models](https://mlsanity.com/paper/2405.20947/) (2024)
- [SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Models](https://mlsanity.com/paper/2402.05044/) (2024)
- [InternLM-XComposer-2.5: A Versatile Large Vision Language Model Supporting Long-Contextual Input and Output](https://mlsanity.com/paper/2407.03320/) (2024)
- [Robust Prompt Optimization for Defending Language Models Against Jailbreaking Attacks](https://mlsanity.com/paper/2401.17263/) (2024)
- [See What You Are Told: Visual Attention Sink in Large Multimodal Models](https://mlsanity.com/paper/2503.03321/) (2025)
- [AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting](https://mlsanity.com/paper/2403.09513/) (2024)
- [Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation](https://mlsanity.com/paper/2403.09572/) (2024)
- [MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance](https://mlsanity.com/paper/2401.02906/) (2024)
- [On Prompt-Driven Safeguarding for Large Language Models](https://mlsanity.com/paper/2401.18018/) (2024)
- [SPA-VL: A Comprehensive Safety Preference Alignment Dataset for Vision Language Model](https://mlsanity.com/paper/2406.12030/) (2024)

## Built on by (most cited first, 4 of 4)

- [Efficient and accurate steering of Large Language Models through attention-guided feature learning](https://mlsanity.com/paper/2602.00333/) (2026)
- [Graph-Regularized Sparse Autoencoders for LLM Safety Steering](https://mlsanity.com/paper/2512.06655/) (2025)
- [Inference-Time Attention Steering for Vision-Language-Action Driving Models](https://mlsanity.com/paper/2608.17095/) (2026)
- [Transfer Safety Awareness for Cross-Modal Safety Drift in Multimodal Large Language Models](https://mlsanity.com/paper/2609.02082/) (2026)

## Abstract (arXiv)

Even modern AI models often remain vulnerable to multimodal queries in which harmful intent is embedded in images. A widely used approach for safety alignment is training with extensive multimodal safety datasets, but the costs of data curation and training are often prohibitive. To mitigate these costs, inference-time alignment has recently been explored, but they often lack generalizability across diverse multimodal jailbreaks and still incur notable overhead due to extra forward passes for response refinement or heavy pre-deployment calibration procedures. Here, we identify insufficient visual attention to safety-critical image regions as one of the key causes of multimodal safety failures. Building on this insight, we propose Multimodal Risk-Adaptive Steering (MoRAS), which enhances safety-critical visual attention via concise visual contexts for accurate multimodal risk assessment. This risk signal enables risk-adaptive steering for direct refusals, reducing inference overhead while remaining generalizable across diverse multimodal jailbreaks. Notably, MoRAS requires only a small calibration set to estimate multimodal risk, substantially reducing pre-deployment overhead. We conduct various empirical validations across multiple benchmarks and MLLM backbones, and observe that the proposed MoRAS consistently mitigates jailbreaks, preserves utility, and reduces computational overhead compared to state-of-the-art inference-time defenses.