mlsanityData to 7 Oct 2026

VLSBench: Unveiling Visual Leakage in Multimodal Safety

arXiv 2411.19939 ↗First posted 29 Nov 2024cs.CR cs.AI cs.CLcited by 59

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

Abstract (arXiv)

Safety concerns of Multimodal large language models (MLLMs) have gradually become an important problem in various applications. Surprisingly, previous works indicate a counterintuitive phenomenon that using textual unlearning to align MLLMs achieves comparable safety performances with MLLMs aligned with image text pairs. To explain such a phenomenon, we discover a Visual Safety Information Leakage (VSIL) problem in existing multimodal safety benchmarks, i.e., the potentially risky content in the image has been revealed in the textual query. Thus, MLLMs can easily refuse these sensitive image-text pairs according to textual queries only, leading to unreliable cross-modality safety evaluation of MLLMs. We also conduct a further comparison experiment between textual alignment and multimodal alignment to highlight this drawback. To this end, we construct multimodal Visual Leakless Safety Bench (VLSBench) with 2.2k image-text pairs through an automated data pipeline. Experimental results indicate that VLSBench poses a significant challenge to both open-source and close-source MLLMs, e.g., LLaVA, Qwen2-VL and GPT-4o. Besides, we empirically compare textual and multimodal alignment methods on VLSBench and find that textual alignment is effective enough for multimodal safety scenarios with VSIL, while multimodal alignment is preferable for safety scenarios without VSIL. Code and data are released under https://github.com/AI45Lab/VLSBench

Key facts. Cited by 59 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 33 of its 56 references that are in the map. First posted on arXiv 29 Nov 2024.

Builds on

33 of 56 references.

Closest ancestors · 6

Same direction, newest first

  1. 2024ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time9 Oct
  2. 2024MOSSBench: Is Your Multimodal Language Model Oversensitive to Safe Queries?22 Jun
  3. 2024Cross-Modal Safety Alignment: Is textual unlearning all you need?27 May
  4. 2024JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks3 Apr
  5. 2024Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation14 Mar
  6. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts9 Nov

Other references · 27

  1. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  2. 2022Training language models to follow instructions with human feedbackcited by 15,712
  3. 2017Proximal Policy Optimization Algorithmscited by 14,441
  4. 2024The Llama 3 Herd of Modelscited by 14,192
  5. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  6. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,151
  7. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,094
  8. 2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,175
  9. 2023InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Taskscited by 2,534
  10. 2023Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyondcited by 2,179
  11. 2023MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGIcited by 2,044
  12. 2024LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Modelscited by 1,442
15 more
  1. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  2. 2023MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilitiescited by 1,221
  3. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,124
  4. 2024MiniCPM-V: A GPT-4V Level MLLM on Your Phonecited by 1,061
  5. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 484
  6. 2024MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmarkcited by 362
  7. 2023DRESS: Instructing Large Vision-Language Models to Align and Interact with Humans via Natural Language Feedbackcited by 99
  8. 2024InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidancecited by 78
  9. 2023Jailbreaking GPT-4V via Self-Adversarial Attacks with System Promptscited by 62
  10. 2024UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Imagescited by 52
  11. 2024Assessment of Multimodal Large Language Models in Alignment with Human Valuescited by 36
  12. 2024From Theft to Bomb-Making: The Ripple Effect of Unlearning in Defending Against Jailbreak Attackscited by 29
  13. 2023GPT-4 as an Effective Zero-Shot Evaluator for Scientific Figure Captionscited by 21
  14. 2024The Dark Side of Function Calling: Pathways to Jailbreaking Large Language Modelscited by 20
  15. 2024OlympicArena Medal Ranks: Who Is the Most Intelligent AI So Far?cited by 10

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 1; 2025: 32; 2026: 26.

59 papers.

Sort:

By area:

  1. 2025A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyondcited by 131
  2. 2025MM-RLHF: The Next Step Forward in Multimodal LLM Alignmentcited by 79
  3. 2024Code-as-Monitor: Constraint-aware Visual Programming for Reactive and Proactive Robotic Failure Detectioncited by 59
  4. 2025IS-Bench: Evaluating Interactive Safety of VLM-Driven Embodied Agents in Daily Household Taskscited by 40
  5. 2025SafeEraser: Enhancing Safety in Multimodal Large Language Models through Multimodal Machine Unlearningcited by 31
  6. 2025Video-SafetyBench: A Benchmark for Safety Evaluation of Video LVLMscited by 25
  7. 2025Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignmentcited by 24
  8. 2025Visual Contextual Attack: Jailbreaking MLLMs with Image-Driven Context Injectioncited by 23
  9. 2025Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Modelscited by 21
  10. 2025X-Boundary: Establishing Exact Safety Boundary to Shield LLMs from Multi-Turn Jailbreaks without Compromising Usabilitycited by 20
  11. 2025Think in Safety: Unveiling and Mitigating Safety Alignment Collapse in Multimodal Large Reasoning Modelcited by 18
  12. 2025Adversary-Aware DPO: Enhancing Safety Alignment in Vision Language Models via Adversarial Trainingcited by 12
38 more
  1. 2025Can AI Perceive Physical Danger and Intervene?cited by 12
  2. 2025Are Your Agents Upward Deceivers?cited by 11
  3. 2025VLMGuard-R1: Proactive Safety Alignment for VLMs via Reasoning-Driven Prompt Optimizationcited by 10
  4. 2025USB: A Comprehensive and Unified Safety Evaluation Benchmark for Multimodal Large Language Modelscited by 10
  5. 2025JPS: Jailbreak Multimodal Large Language Models with Collaborative Visual Perturbation and Textual Steeringcited by 10
  6. 2025SEA: Low-Resource Safety Alignment for Multimodal Large Language Models via Synthetic Embeddingscited by 9
  7. 2025Scaling Beyond Context: A Survey of Multimodal Retrieval-Augmented Generation for Document Understandingcited by 9
  8. 2025Speech-Audio Compositional Attacks on Multimodal LLMs and Their Mitigation with SALMONN-Guardcited by 8
  9. 2025LED-Merging: Mitigating Safety-Utility Conflicts in Model Merging with Location-Election-Disjointcited by 7
  10. 2025LLMs Deceive Unintentionally: Emergent Misalignment in Dishonesty from Misaligned Samples to Biased Human-AI Interactionscited by 7
  11. 2025VLSU: Mapping the Limits of Joint Multimodal Understanding for AI Safetycited by 7
  12. 2025HoliSafe: Holistic Safety Benchmarking and Modeling for Vision-Language Modelcited by 6
  13. 2026Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarkscited by 6
  14. 2025ShieldVLM: Safeguarding the Multimodal Implicit Toxicity via Deliberative Reasoning with LVLMscited by 5
  15. 2025Implicit Jailbreak Attacks via Cross-Modal Information Concealment on Vision-Language Modelscited by 5
  16. 2025REVEAL: Multi-turn Evaluation of Image-Input Harms for Vision LLMcited by 4
  17. 2026SAVeS: Steering Safety Judgments in Vision-Language Models via Semantic Cuescited by 4
  18. 2025Beyond External Monitors: Enhancing Transparency of Large Language Models for Easier Monitoringcited by 2
  19. 2025Plane Geometry Problem Solving with Multi-modal Reasoning: A Surveycited by 2
  20. 2025Self-adaptive Dataset Construction for Real-World Multimodal Safety Scenarioscited by 2
  21. 2025Chameleon: Adaptive Adversarial Agents for Scaling-Based Visual Prompt Injection in Multimodal AI Systemscited by 2
  22. 2026When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily Lifecited by 2
  23. 2026CSR-Bench: A Benchmark for Evaluating the Cross-modal Safety and Reliability of MLLMscited by 2
  24. 2026Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Imagescited by 2
  25. 2026Multimodal Reward Hacking in Reinforcement Learningcited by 2
  26. 2026Automated Safety Benchmarking: A Multi-agent Pipeline for LVLMscited by 1
  27. 2026EviSafe: Evidence-Grounded Safety Evaluation for Vision-Language Modelscited by 1
  28. 2025SurvSurf: a partially monotonic neural network for first-hitting time prediction of intermittently observed discrete and continuous sequential eventscited by 0
  29. 2025Visual Attention Reasoning via Hierarchical Search and Self-Verificationcited by 0
  30. 2026Towards Multi-modal Multi-turn Safety: From Agentic Interaction to Strategic Alignmentcited by 0
  31. 2026Unveiling Covert Toxicity in Multimodal Data via Toxicity Association Graphs: A Graph-Based Metric and Interpretable Detection Frameworkcited by 0
  32. 2026Controlling Chat Style in Language Models via Single-Direction Editingcited by 0
  33. 2026OOD-MMSafe: Advancing MLLM Safety from Harmful Intent to Hidden Consequencescited by 0
  34. 2026SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamicscited by 0
  35. 2026Jailbreaking Multimodal Large Language Models using Multi-Clip Videocited by 0
  36. 2026Constitutional On-Policy Safe Distillationcited by 0
  37. 2026Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gatingcited by 0
  38. 2026LoMC: Localized Multidirectional Correction for Refusal Suppression in Routed Foundation Modelscited by 0

50 most cited of 59 All 59 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs
  2. 2025SDEval: Safety Dynamic Evaluation for Multimodal Large Language Models
  3. 2024Safety of Multimodal Large Language Models on Images and Texts
  4. 2023MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models
  5. 2025VSCBench: Bridging the Gap in Vision-Language Model Safety Calibration
  6. 2026Lingua-SafetyBench: A Benchmark for Safety Evaluation of Multilingual Vision-Language Models
  7. 2024CoCA: Regaining Safety-awareness of Multimodal Large Language Models with Constitutional Calibration
  8. 2025Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models?