Data to 7 Oct 2026

MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance

arXiv 2401.02906 ↗First posted 5 Jan 2024cs.CR cs.CL cs.CVcited by 111

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness

Abstract (arXiv)

The deployment of multimodal large language models (MLLMs) has brought forth a unique vulnerability: susceptibility to malicious attacks through visual inputs. This paper investigates the novel challenge of defending MLLMs against such attacks. Compared to large language models (LLMs), MLLMs include an additional image modality. We discover that images act as a ``foreign language" that is not considered during safety alignment, making MLLMs more prone to producing harmful responses. Unfortunately, unlike the discrete tokens considered in text-based LLMs, the continuous nature of image signals presents significant alignment challenges, which poses difficulty to thoroughly cover all possible scenarios. This vulnerability is exacerbated by the fact that most state-of-the-art MLLMs are fine-tuned on limited image-text pairs that are much fewer than the extensive text-based pretraining corpus, which makes the MLLMs more prone to catastrophic forgetting of their original abilities during safety fine-tuning. To tackle these challenges, we introduce MLLM-Protector, a plug-and-play strategy that solves two subtasks: 1) identifying harmful responses via a lightweight harm detector, and 2) transforming harmful responses into harmless ones via a detoxifier. This approach effectively mitigates the risks posed by malicious visual inputs without compromising the original performance of MLLMs. Our results demonstrate that MLLM-Protector offers a robust solution to a previously unaddressed aspect of MLLM security.

Key facts. Cited by 111 papers in the map (data through 7 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 44 of its 63 references that are in the map. First posted on arXiv 5 Jan 2024.

Builds on

44 of 63 references.

Closest ancestors · 2

Same direction, newest first

Other references · 42

2020Language Models are Few-Shot Learnerscited by 32,199
2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
2022Training language models to follow instructions with human feedbackcited by 15,712
2017Proximal Policy Optimization Algorithmscited by 14,441
2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
2022PaLM: Scaling Language Modeling with Pathwayscited by 5,304
2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
2022Constitutional AI: Harmlessness from AI Feedbackcited by 2,950
2020Learning to summarize from human feedbackcited by 2,562
2023Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyondcited by 2,179
2019Fine-Tuning Language Models from Human Preferencescited by 2,124

30 more

2022BLOOM: A 176B-Parameter Open-Access Multilingual Language Modelcited by 1,968
2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,734
2021WebGPT: Browser-assisted question-answering with human feedbackcited by 1,722
2023MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilitiescited by 1,221
2022Ignore Previous Prompt: Attack Techniques For Language Modelscited by 760
2023RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignmentcited by 672
2023Safe RLHF: Safe Reinforcement Learning from Human Feedbackcited by 630
2023LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Modelcited by 602
2023Prompt Injection attack against LLM-integrated Applicationscited by 568
2022Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Modelcited by 557
2022Improving alignment of dialogue agents via targeted human judgementscited by 527
2023RRHF: Rank Responses to Align Language Models with Human Feedback without tearscited by 454
2018Scalable agent alignment via reward modeling: a research directioncited by 438
2023SmoothLLM: Defending Large Language Models Against Jailbreaking Attackscited by 434
2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrationscited by 401
2018AI safety via debatecited by 384
2023PandaGPT: One Model To Instruction-Follow Them Allcited by 330
2023Statistical Rejection Sampling Improves Preference Optimizationcited by 321
2021Recursively Summarizing Books with Human Feedbackcited by 287
2023Exploiting Programmatic Behavior of LLMs: Dual-Use Through Standard Security Attackscited by 268
2023Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Modelscited by 237
2022ZeroGen: Efficient Zero-shot Learning via Dataset Generationcited by 209
2023Mitigating the Alignment Tax of RLHFcited by 181
2018Supervising strong learners by amplifying weak expertscited by 139
2023Understanding Catastrophic Forgetting in Language Models via Implicit Inferencecited by 138
2023Training Language Models with Language Feedback at Scalecited by 125
2023DetGPT: Detect What You Need via Reasoningcited by 112
2023LMFlow: An Extensible Toolkit for Finetuning and Inference of Large Foundation Modelscited by 58
2023HiLM-D: Enhancing MLLMs with Multi-Scale High-Resolution Details for Autonomous Drivingcited by 48
2023Gaining Wisdom from Setbacks: Aligning Large Language Models via Mistake Analysiscited by 34

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 32; 2025: 51; 2026: 28.

111 papers.

Sort:

By area:

2024The Revolution of Multimodal Large Language Models: A Surveycited by 134
2024Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformationcited by 114
2024AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Promptingcited by 113
2024Jailbreak Vision Language Models via Bi-Modal Adversarial Promptcited by 103
2024SPA-VL: A Comprehensive Safety Preference Alignment Dataset for Vision Language Modelcited by 82
2024A Survey of Attacks on Large Vision-Language Models: Resources, Advances, and Future Trendscited by 82
2024Safety of Multimodal Large Language Models on Images and Textscited by 70
2024JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Modelscited by 67
2025A Survey of Safety on Large Vision-Language Models: Attacks, Defenses and Evaluationscited by 66
2024CARES: A Comprehensive Benchmark of Trustworthiness in Medical Vision Language Modelscited by 64
2024ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Timecited by 57
2024Jailbreak Large Vision-Language Models Through Multi-Modal Linkagecited by 53

38 more

2024Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreakscited by 44
2024Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignmentcited by 42
2025Backdoor Cleaning without External Guidance in MLLM Fine-tuningcited by 34
2025VLM2-Bench: A Closer Look at How Well VLMs Implicitly Link Explicit Matching Visual Cuescited by 33
2024Jailbreak Attacks and Defenses against Multimodal Generative Models: A Surveycited by 31
2025MMUnlearner: Reformulating Multimodal Machine Unlearning in the Era of Multimodal Large Language Modelscited by 29
2025HiddenDetect: Detecting Jailbreak Attacks against Large Vision-Language Models via Monitoring Hidden Statescited by 29
2025Safe RLHF-V: Safe Reinforcement Learning from Multi-modal Human Feedbackcited by 27
2024The First to Know: How Token Distributions Reveal Hidden Knowledge in Large Vision-Language Models?cited by 24
2024Safe Inputs but Unsafe Output: Benchmarking Cross-modality Safety Alignment of Large Vision-Language Modelcited by 24
2024Cross-modality Information Check for Detecting Jailbreaking in Multimodal Large Language Modelscited by 24
2024Personalized Visual Instruction Tuningcited by 24
2024Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Modelscited by 23
2025Visual Contextual Attack: Jailbreaking MLLMs with Image-Driven Context Injectioncited by 23
2024Mission Impossible: A Statistical Perspective on Jailbreaking LLMscited by 21
2024The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defensecited by 20
2024RapGuard: Safeguarding Multimodal Large Language Models via Rationale-aware Defensive Promptingcited by 19
2025A Survey on Agentic Multimodal Large Language Modelscited by 19
2024Direct Large Language Model Alignment Through Self-Rewarding Contrastive Prompt Distillationcited by 18
2024Unbridled Icarus: A Survey of the Potential Perils of Image Inputs in Multimodal Large Language Model Securitycited by 18
2024UniGuard: Towards Universal Safety Guardrails for Jailbreak Attacks on Multimodal Large Language Modelscited by 18
2024VLMGuard: Bootstrapping Malicious Prompt Detectors from Unlabeled Vision-Language Prompts in the Wildcited by 17
2025Understanding and Rectifying Safety Perception Distortion in VLMscited by 17
2025Safety Mirage: How Spurious Correlations Undermine VLM Safety Fine-Tuning and Can Be Mitigated by Machine Unlearningcited by 16
2024The Synergy between Data and Multi-Modal Large Language Models: A Survey from Co-Development Perspectivecited by 14
2025Survey of Adversarial Robustness in Multimodal Large Language Modelscited by 14
2024The Instinctive Bias: Spurious Images lead to Illusion in MLLMscited by 11
2024Jailbreaking as a Reward Misspecification Problemcited by 11
2025Unlearning Sensitive Information in Multimodal LLMs: Benchmark and Attack-Defense Evaluationcited by 11
2025SafeGRPO: Self-Rewarded Multimodal Safety Alignment via Rule-Governed Policy Optimizationcited by 11
2025VLMGuard-R1: Proactive Safety Alignment for VLMs via Reasoning-Driven Prompt Optimizationcited by 10
2025OmniSafeBench-MM: A Unified Benchmark and Toolbox for Multimodal Jailbreak Attack-Defense Evaluationcited by 10
2025Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models?cited by 9
2025SafePTR: Token-Level Jailbreak Defense in Multimodal LLMs via Prune-then-Restore Mechanismcited by 9
2025Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoringcited by 9
2025PiCo: Jailbreaking Multimodal Large Language Models via Pictorial Code Contextualizationcited by 8
2025SaFeR-VLM: Toward Safety-aware Fine-grained Reasoning in Multimodal Modelscited by 8
2024Defending LVLMs Against Vision Attacks through Partial-Perception Supervisioncited by 7

50 most cited of 111 All 111 (JSON)

Similar, not cited

Close in wording, no citation either way