mlsanityData to 7 Oct 2026

Vision-language and text-to-image safety

507 papers2 this weekLow confidence in this grouping: the papers may belong to more than one topic, or the group may changeLow confidence in this grouping: the papers may belong to more than one topic, or the group may change · About

Keywords: safety, visual, multimodal, adversarial, diffusion

Key facts. Vision-language and text-to-image safety: 507 papers on arXiv since 2022 (data through 7 Oct 2026). 183 of its 507 mapped papers (36%) cite “FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts” (2023, arXiv 2311.05608). 117 of its 507 mapped papers (23%) cite “Visual Adversarial Examples Jailbreak Aligned Large Language Models” (2023, arXiv 2306.13213). 13 new papers 24 Sep–7 Oct.

What’s new

24 Sep – 7 Oct 2026

13 new papers, 24 Sep – 7 Oct; 2 this week.

  1. This week · 2
  2. Sparse Feature Policy Unlearning Mitigates State Hallucination in Vision-Language-Action Models
  3. Harmful Content Generation in Text-to-Image Models: Capabilities and Moderation Limitations
  4. Earlier · 11
  5. FADE: Frame-Aware Diffusion-Transformer-based Multi-Concept Erasure for Video Unlearning also fits: Machine unlearning
  6. Walking the Embedding Space: Datastore Extraction from Multimodal RAG
  7. ShieldCLIP: Selective Safety Alignment for Harmful Content Mitigation in Multimodal Foundation Models also fits: Safety alignment, refusal and guardrails
  8. FOMO: Forget the Concept, Don't Miss Out on the Scene in Selective Video Unlearning
  9. Steering Fields: Adaptive Vector Fields for Safe Image Generation and Beyond
  10. CollageAttack: Exploiting Cross-Modal Alignment Flaws in T2I Models through Spatial Text Composition
  11. Selective Channel Restoration for Backdoored Vision-Language Models
  12. Motion Concept Unlearning in Video Diffusion Models
All 13
  1. Render Before Reading: Visual Rendering as a Prompt Injection Defense
  2. From Latents to Wires: Surgical Post-Editing on Large Language Models
  3. Region-Level Black-Box Defense Against Stealthy Embedding-Space Backdoors in CLIP also fits: Backdoor and data poisoning attacks

Foundations

Most cited by papers here

  1. FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts2023cited by 183 here
  2. Visual Adversarial Examples Jailbreak Aligned Large Language Models2023cited by 117 here
  3. Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models2023cited by 110 here
  4. Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?2023cited by 106 here
  5. Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Models2023cited by 93 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 61 here
  2. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 136 here
  3. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 65 here
  4. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Promptscited by 183 here
  5. 2024Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Modelscited by 60 here
  6. 2024AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Promptingcited by 70 here
  7. 2024JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attackscited by 68 here
17 more it builds on, not in this chain
  1. 2022Red-Teaming the Stable Diffusion Safety Filtercited by 89 here
  2. 2023Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Modelscited by 93 here
  3. 2023SneakyPrompt: Jailbreaking Text-to-image Generative Modelscited by 79 here
  4. 2023Visual Adversarial Examples Jailbreak Aligned Large Language Modelscited by 117 here
  5. 2023Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Modelscited by 110 here
  6. 2023Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Promptscited by 67 here
  7. 2023AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Modelscited by 43 here
  8. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 39 here
  9. 2023Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?cited by 106 here
  10. 2023To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Nowcited by 79 here
  11. 2023SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generationcited by 74 here
  12. 2024Jailbreaking Attack against Multimodal Large Language Modelcited by 62 here
  13. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 50 here
  14. 2024Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Modelscited by 81 here
  15. 2024Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Modelscited by 67 here
  16. 2024Jailbreak Vision Language Models via Bi-Modal Adversarial Promptcited by 41 here
  17. 2024SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generationcited by 46 here

The 24 most cited of 991.

Browse

30 of 507, ranked 6–35

  1. 6Red-Teaming the Stable Diffusion Safety Filter202289
  2. 7Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models202481
  3. 8SneakyPrompt: Jailbreaking Text-to-image Generative Models202379
  4. 9To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now202379
  5. 10AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting202470
  6. 11Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models202467
  7. 12Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts202367
  8. 13Jailbreaking Attack against Multimodal Large Language Model202462
  9. 14Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models202460
  10. 15SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation202446
All 30
  1. 16Jailbreak Vision Language Models via Bi-Modal Adversarial Prompt202441
  2. 17On the Adversarial Robustness of Multi-Modal Foundation Models202337
  3. 18Jailbreak Large Vision-Language Models Through Multi-Modal Linkage202436
  4. 19White-box Multimodal Jailbreaks Against Large Vision-Language Models202433
  5. 20Jailbreaking Prompt Attack: A Controllable Adversarial Attack against Diffusion Models202431
  6. 21Visual-RolePlay: Universal Jailbreak Attack on MultiModal Large Language Models via Role-playing Image Character202430
  7. 22SafeGen: Mitigating Sexually Explicit Content Generation in Text-to-Image Models202429
  8. 23Jailbreaking Multimodal Large Language Models via Shuffle Inconsistency202528
  9. 24Perception-guided Jailbreak against Text-to-Image Models202428
  10. 25Distraction is All You Need for Multimodal Large Language Model Jailbreaking202527
  11. 26SPA-VL: A Comprehensive Safety Preference Alignment Dataset for Vision Language Model202426
  12. 27UnlearnCanvas: Stylized Image Dataset for Enhanced Machine Unlearning Evaluation in Diffusion Models202425
  13. 28Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient202425
  14. 29GuardT2I: Defending Text-to-Image Models from Adversarial Prompts202423
  15. 30ImgTrojan: Jailbreaking Vision-Language Models with ONE Image202422
  16. 31Direct Unlearning Optimization for Robust and Safe Text-to-Image Models202422
  17. 32ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users202421
  18. 33A Survey of Attacks on Large Vision-Language Models: Resources, Advances, and Future Trends202421
  19. 34ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning202420
  20. 35IDEATOR: Jailbreaking and Benchmarking Large Vision-Language Models Using Themselves202420