SafeEraser: Enhancing Safety in Multimodal Large Language Models through Multimodal Machine Unlearning

Challenge:	Existing methods for MU forget quality and model utility are not fully explored for safety in MLLMs.
Approach:	They propose a safety unlearning benchmark for MLLMs to measure over-forgetting . they propose MU methods to forget quality and model utility .
Outcome:	The proposed method reduces over-forgetting by 79.5% while maintaining forget quality and model utility.

Similar Papers

Overview

Challenge

Approach

Outcome

Protecting Privacy in Multimodal Large Language Models with MLLMU-Bench (2025.naacl-long)

Copied to clipboard

Zheyuan Liu, Guangyao Dou, Mengzhao Jia, Zhaoxuan Tan, Qingkai Zeng, Yongle Yuan, Meng Jiang

Language Models

Challenge:	Large Language Models (LLMs) and Multimodal Large Language models (MLLMs) trained on vast web corpora can memorize and disclose individuals’ confidential and private data, raising legal and ethical concerns.
Approach:	They propose a benchmark to assess unlearning algorithms from multiple perspectives and provide a baseline for existing generative models.
Outcome:	The proposed benchmark consists of 500 fictitious profiles and 153 profiles for public celebrities, evaluated from both multimodal (image+text) and unimodal (text) perspectives.

OFFSIDE: Benchmarking Unlearning Misinformation in Multimodal Large Language Models (2026.findings-acl)

Copied to clipboard

Hao Zheng, Zirui Pang, Ling Li, Zhijie Deng, Yuhan Pu, Zhaowei Zhu, Xiaobo Xia, Jiaheng Wei

Responsible & Trustworthy NLP

Challenge:	Existing benchmarks for MU are limited by a lack of image diversity and coarse-grained unlearning targets.
Approach:	They propose a benchmark to evaluate misinformation unlearning in MLLMs . OFFSIDE supports advanced unlearning targets such as fine-grained unlearning and visual rumor removal.
Outcome:	OFFSIDE supports advanced unlearning targets, such as fine-grained unlearning and visual rumor removal.

SUA: Stealthy Multimodal Large Language Model Unlearning Attack (2025.emnlp-main)

Copied to clipboard

Xianren Zhang, Hui Liu, Delvin Ce Zhang, Xianfeng Tang, Qi He, Dongwon Lee, Suhang Wang

Semantic Text Processing

Challenge:	Multimodal Large Language Models (MLLMs) trained on massive data may memorize sensitive personal information and photos, posing privacy and copyright concerns.
Approach:	They propose a framework that learns a universal noise pattern to recover unlearned information from MLLMs.
Outcome:	The proposed framework learns a universal noise pattern and can reveal unlearned content when applied to images.

SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models (2026.findings-acl)

Copied to clipboard

Xinyi Zeng, Xue Yang, Jingyuan Zhang, Huanqian Yan, Xiang Chen, Kaiwen Wei, Hankun Kang, Yu Tian

Responsible & Trustworthy NLP

Challenge:	Existing defense methods rely on fine-tuning or inefficient post-hoc interventions, limiting their ability to address novel attacks.
Approach:	They propose a decoding-level defense mechanism that employs a lightweight discriminator to iteratively steer the decoding process toward safety.
Outcome:	The proposed method improves safety performance by up to 33.40% without fine-tuning on multiple MLLMs.

Can’t See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs (2025.acl-long)

Copied to clipboard

Wenxuan Wang, Xiaoyuan Liu, Kuiyi Gao, Jen-tse Huang, Youliang Yuan, Pinjia He, Shuai Wang, Zhaopeng Tu

Language Models

Challenge:	Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images.
Approach:	They propose a multimodal safety awareness benchmark to evaluate MLLMs across 29 safety scenarios with 1,500 carefully curated image-prompt pairs.
Outcome:	The proposed model is able to identify unsafe content and avoid over-sensitivity that can hinder helpfulness.

SURE: Safety Understanding and Reasoning Enhancement for Multimodal Large Language Models (2025.emnlp-main)

Copied to clipboard

Yuxin Gou, Xiaoning Dong, Qin Li, Shishen Gu, Richang Hong, Wenbo Hu

Language Models

Challenge:	Existing multimodal large language models incorporate visual and textual information, but introduces new and complex safety risks.
Approach:	They propose a safety reasoning framework that integrates visual modalities into multimodal models to help them resist jailbreak attacks.
Outcome:	The proposed framework improves model safety while avoiding over-defense . it is based on a large-scale safety reasoning dataset .

MLLM-Protector: Ensuring MLLM’s Safety without Hurting Performance (2024.emnlp-main)

Copied to clipboard

Renjie Pi, Tianyang Han, Jianshu Zhang, Yueqi Xie, Rui Pan, Qing Lian, Hanze Dong, Jipeng Zhang, Tong Zhang

Responsible & Trustworthy NLP

Challenge:	MLLMs are deployed on limited image-text pairs, which makes them more vulnerable to catastrophic forgetting of their original abilities during safety fine-tuning.
Approach:	They propose a plug-and-play strategy that detects harmful visual inputs and transforms harmful ones into harmless ones.
Outcome:	The proposed approach mitigates the risks posed by malicious visual inputs without compromising the original performance of MLLMs.

MMUnlearner: Reformulating Multimodal Machine Unlearning in the Era of Multimodal Large Language Models (2025.findings-acl)

Copied to clipboard

Jiahao Huo, Yibo Yan, Xu Zheng, Yuanhuiyi Lyu, Xin Zou, Zhihua Wei, Xuming Hu

Semantic Text Processing

Challenge:	Recent advances in machine learning (MU) have enabled the selective removal of private or sensitive information encoded within deep neural networks.
Approach:	They propose to "reformulate" the task of multimodal MU in the era of MLLMs by preserving only the visual patterns associated with a given entity while preserving the corresponding textual knowledge.
Outcome:	The proposed method surpasses baselines that finetuned MLLMs with VQA data directly through Gradient Ascent (GA) or Negative Preference Optimization (NPO), across all evaluation dimensions.

Unveiling Uncertainty: A Deep Dive into Calibration and Performance of Multimodal Large Language Models (2025.coling-main)

Copied to clipboard

Zijun Chen, Wenbo Hu, Guande He, Zhijie Deng, ZHeng ZHang, Richang Hong

Multimodality

Challenge:	Multimodal large language models combine visual and textual data for tasks like image captioning and visual question answering.
Approach:	They propose temperature scaling and iterative prompt optimization to calibrate MLLMs and enhance model reliability.
Outcome:	The proposed techniques improve MLLMs and improve model reliability.

Automating Steering for Safe Multimodal Large Language Models (2025.emnlp-main)

Copied to clipboard

Lyucheng Wu, Mengru Wang, Ziwen Xu, Tri Cao, Nay Oo, Bryan Hooi, Shumin Deng

Language Models

Challenge:	Recent advances in Multimodal Large Language Models (MLLMs) have unlocked powerful cross-modal reasoning abilities, but also raised new safety concerns, especially when faced with adversarial multimodal inputs.
Approach:	They propose a modular and adaptive inference-time intervention technology, AutoSteer, that integrates a safety awareness score, an adaptive safety prober, and a lightweight Refusal Head to modulate generation when safety risks are detected.
Outcome:	Experiments on LLaVA-OV and Chameleon show that AutoSteer significantly reduces the Attack Success Rate (ASR) for textual, visual, and cross-modal threats while maintaining general abilities.