# Layer-wise Alignment: Examining Safety Alignment Across Image Encoder Layers in Vision Language Models

arXiv: https://arxiv.org/abs/2411.04291 · HTML: https://mlsanity.com/paper/2411.04291/

## Key facts

- Cited by 4 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 48 of its 65 references that are in the map.
- First posted on arXiv 6 Nov 2024.

## 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/)
- Efficient LLM inference and serving / [Visual token pruning in multimodal models](https://mlsanity.com/inference/visual-token-pruning-in-multimodal-models/)
- 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 48)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [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)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Explaining and Harnessing Adversarial Examples](https://mlsanity.com/paper/1412.6572/) (2014)
- [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://mlsanity.com/paper/2201.12086/) (2022)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Towards Evaluating the Robustness of Neural Networks](https://mlsanity.com/paper/1608.04644/) (2016)
- [Trust Region Policy Optimization](https://mlsanity.com/paper/1502.05477/) (2015)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [Jailbroken: How Does LLM Safety Training Fail?](https://mlsanity.com/paper/2307.02483/) (2023)
- [DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing](https://mlsanity.com/paper/2111.09543/) (2021)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [Constrained Policy Optimization](https://mlsanity.com/paper/1705.10528/) (2017)
- [Adversarial Examples Are Not Bugs, They Are Features](https://mlsanity.com/paper/1905.02175/) (2019)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [RLHF-V: Towards Trustworthy MLLMs via Behavior Alignment from Fine-grained Correctional Human Feedback](https://mlsanity.com/paper/2312.00849/) (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)
- [Large Language Model Unlearning](https://mlsanity.com/paper/2310.10683/) (2023)
- [Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models](https://mlsanity.com/paper/2307.14539/) (2023)
- [RLHF Workflow: From Reward Modeling to Online RLHF](https://mlsanity.com/paper/2405.07863/) (2024)
- [JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks](https://mlsanity.com/paper/2404.03027/) (2024)
- [Aligning Modalities in Vision Large Language Models via Preference Fine-tuning](https://mlsanity.com/paper/2402.11411/) (2024)

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

- [A Survey of Safety on Large Vision-Language Models: Attacks, Defenses and Evaluations](https://mlsanity.com/paper/2502.14881/) (2025)
- [Understanding Language Prior of LVLMs by Contrasting Chain-of-Embedding](https://mlsanity.com/paper/2509.23050/) (2025)
- [STARE: Step-wise Temporal Alignment and Red-teaming Engine for Multi-modal Toxicity Attack](https://mlsanity.com/paper/2605.00699/) (2026)
- [New Wide-Net-Casting Jailbreak Attacks Risk Large Models](https://mlsanity.com/paper/2605.17128/) (2026)

## Abstract (arXiv)

Vision-language models (VLMs) have improved significantly in their capabilities, but their complex architecture makes their safety alignment challenging. In this paper, we reveal an uneven distribution of harmful information across the intermediate layers of the image encoder and show that skipping a certain set of layers and exiting early can increase the chance of the VLM generating harmful responses. We call it as "Image enCoder Early-exiT" based vulnerability (ICET). Our experiments across three VLMs: LLaVA-1.5, LLaVA-NeXT, and Llama 3.2, show that performing early exits from the image encoder significantly increases the likelihood of generating harmful outputs. To tackle this, we propose a simple yet effective modification of the Clipped-Proximal Policy Optimization (Clip-PPO) algorithm for performing layer-wise multi-modal RLHF for VLMs. We term this as Layer-Wise PPO (L-PPO). We evaluate our L-PPO algorithm across three multimodal datasets and show that it consistently reduces the harmfulness caused by early exits.