# Activation Steering Meets Preference Optimization: Defense Against Jailbreaks in Vision Language Models

arXiv: https://arxiv.org/abs/2509.00373 · HTML: https://mlsanity.com/paper/2509.00373/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 36 of its 42 references that are in the map.
- First posted on arXiv 30 Aug 2025.

## Placed in

- 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/)
- LLM safety and security / [Vision-language and text-to-image safety](https://mlsanity.com/safety/vision-language-and-text-to-image-safety/)
- Mechanistic interpretability / [Steering and probing for LLM safety](https://mlsanity.com/mechinterp/steering-and-probing-for-llm-safety/)

## Builds on (most cited first, 30 of 36)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [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)
- [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://mlsanity.com/paper/2305.06500/) (2023)
- [MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models](https://mlsanity.com/paper/2304.10592/) (2023)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilities](https://mlsanity.com/paper/2308.02490/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models](https://mlsanity.com/paper/2310.04451/) (2023)
- [The Linear Representation Hypothesis and the Geometry of Large Language Models](https://mlsanity.com/paper/2311.03658/) (2023)
- [CogVLM: Visual Expert for Pretrained Language Models](https://mlsanity.com/paper/2311.03079/) (2023)
- [MiniGPT-v2: large language model as a unified interface for vision-language multi-task learning](https://mlsanity.com/paper/2310.09478/) (2023)
- [GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Prompts](https://mlsanity.com/paper/2309.10253/) (2023)
- [FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts](https://mlsanity.com/paper/2311.05608/) (2023)
- [Visual Adversarial Examples Jailbreak Aligned Large Language Models](https://mlsanity.com/paper/2306.13213/) (2023)
- [Are aligned neural networks adversarially aligned?](https://mlsanity.com/paper/2306.15447/) (2023)
- [MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models](https://mlsanity.com/paper/2311.17600/) (2023)
- [Extracting Latent Steering Vectors from Pretrained Language Models](https://mlsanity.com/paper/2205.05124/) (2022)
- [COLD-Attack: Jailbreaking LLMs with Stealthiness and Controllability](https://mlsanity.com/paper/2402.08679/) (2024)
- [On the Adversarial Robustness of Multi-Modal Foundation Models](https://mlsanity.com/paper/2308.10741/) (2023)
- [Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models](https://mlsanity.com/paper/2403.09792/) (2024)
- [Jailbreaking Attack against Multimodal Large Language Model](https://mlsanity.com/paper/2402.02309/) (2024)
- [AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting](https://mlsanity.com/paper/2403.09513/) (2024)

## Built on by (most cited first, 5 of 5)

- [Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models](https://mlsanity.com/paper/2603.19233/) (2026)
- [Efficient and accurate steering of Large Language Models through attention-guided feature learning](https://mlsanity.com/paper/2602.00333/) (2026)
- [Diagnosing and Repairing Unsafe Channels in Vision-Language Models via Causal Discovery and Dual-Modal Safety Subspace Projection](https://mlsanity.com/paper/2603.27240/) (2026)
- [ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models](https://mlsanity.com/paper/2605.15687/) (2026)
- [Do GUI Agents Know When Not to Act? Enabling Conflict-Aware Termination for Multimodal GUI Agents](https://mlsanity.com/paper/2609.03438/) (2026)

## Abstract (arXiv)

Vision Language Models (VLMs) have demonstrated impressive capabilities in integrating visual and textual information for understanding and reasoning, but remain highly vulnerable to adversarial attacks. While activation steering has emerged as a promising defence, existing approaches often rely on task-specific contrastive prompts to extract harmful directions, which exhibit suboptimal performance and can degrade visual grounding performance. To address these limitations, we propose textitSequence-Level Preference Optimization for VLM (textitSPO-VLM), a novel two-stage defense framework that combines activation-level intervention with policy-level optimization to enhance model robustness. In textitStage I, we compute adaptive layer-specific steering vectors from diverse data sources, enabling generalized suppression of harmful behaviors during inference. In textitStage II, we refine these steering vectors through a sequence-level preference optimization process. This stage integrates automated toxicity assessment, as well as visual-consistency rewards based on caption-image alignment, to achieve safe and semantically grounded text generation. The two-stage structure of SPO-VLM balances efficiency and effectiveness by combining a lightweight mitigation foundation in Stage I with deeper policy refinement in Stage II. Extensive experiments shown SPO-VLM enhances safety against attacks via activation steering and preference optimization, while maintaining strong performance on benign tasks without compromising visual understanding capabilities. We will release our code, model weights, and evaluation toolkit to support reproducibility and future research. textcolorredWarning: This paper may contain examples of offensive or harmful text and images.