mlsanityData to 14 Oct 2026

DIVA: Exploiting Cross-Step Conditional Propagation for Visual Jailbreaks in Discrete Diffusion Vision-Language Models

arXiv 2609.05525 ↗First posted 1 Sep 2026cs.CVcited by 0

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

also in 2 more
Abstract (arXiv)

Large vision-language models (VLMs) are increasingly deployed in safety-critical settings, yet existing visual jailbreak research has focused almost exclusively on autoregressive architectures, leaving an important emerging family unstudied: multimodal discrete diffusion vision-language models (dVLMs). We identify a vulnerability specific to diffusion generation: because the visual embedding conditions every reverse denoising step rather than acting as a one-time prefix, adversarial visual semantics are repeatedly propagated and amplified across the generation trajectory, a phenomenon we term cross-step conditional propagation. We provide empirical evidence through stage-sensitivity analysis, prompt-level switch rates, and pairwise denoising-bin disagreement metrics, confirmed by bootstrap resampling. We propose DIVA (Discrete-diffusion Vision-language model Attack), a white-box visual jailbreak framework using cross-modal intent obfuscation and diffusion-aware multi-timestep adversarial optimization. Across three dVLMs, DIVA reaches 58.8%, 67.7%, and 69.1% HADES ASR under the Beaver reward-model metric, outperforming visual jailbreak baselines designed for autoregressive models. Code: https://github.com/loststars2002/DIVA

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 36 of its 37 references that are in the map. First posted on arXiv 1 Sep 2026.

Builds on

36 of 37 references.

Closest ancestors · 15

Same direction, newest first

  1. 2026Adversarial Attacks on Multimodal Large Language Models: A Comprehensive Survey30 Mar
  2. 2026Jailbreaks on Vision Language Model via Multimodal Reasoning29 Jan
  3. 2025Jailbreaking Large Vision Language Models in Intelligent Transportation Systems17 Nov
  4. 2025ARMs: Adaptive Red-Teaming Agent against Multimodal Models with Plug-and-Play Attacks3 Oct
  5. 2025CAVALRY-V: A Large-Scale Generator Framework for Adversarial Attacks on Video MLLMs1 Jul
  6. 2025Video-SafetyBench: A Benchmark for Safety Evaluation of Video LVLMs17 May
  7. 2024VLSBench: Unveiling Visual Leakage in Multimodal Safety29 Nov
  8. 2024Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignment27 Nov
  9. 2024A Survey of Attacks on Large Vision-Language Models: Resources, Advances, and Future Trends10 Jul
  10. 2024Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models14 Mar
  11. 2024Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation14 Mar
  12. 2023MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models29 Nov
3 more
  1. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts9 Nov
  2. 2023Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models26 Jul
  3. 2023Visual Adversarial Examples Jailbreak Aligned Large Language Models22 Jun

Other references · 21

  1. 2023Visual Instruction Tuningcited by 7,869
  2. 2023Improved Baselines with Visual Instruction Tuningcited by 4,418
  3. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  4. 2025Qwen3-VL Technical Reportcited by 2,793
  5. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
  6. 2025Large Language Diffusion Modelscited by 912
  7. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 860
  8. 2024A StrongREJECT for Empty Jailbreakscited by 385
  9. 2025LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Modelscited by 285
  10. 2025MMaDA: Multimodal Large Diffusion Language Modelscited by 250
  11. 2025LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuningcited by 148
  12. 2025LaViDa: A Large Diffusion Language Model for Multimodal Understandingcited by 84
9 more
  1. 2025The Devil behind the mask: An emergent safety vulnerability of Diffusion LLMscited by 34
  2. 2025Jailbreaking Large Language Diffusion Models: Revealing Hidden Safety Flaws in Diffusion-Based Text Generationcited by 26
  3. 2026LLaDA2.0-Uni: Unifying Multimodal Understanding and Generation with Diffusion Large Language Modelcited by 13
  4. 2025Where to Start Alignment? Diffusion Large Language Model May Demand a Distinct Positioncited by 11
  5. 2025Toward Safer Diffusion Language Models: Discovery and Mitigation of Priming Vulnerabilitycited by 10
  6. 2026LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Modelcited by 10
  7. 2026Safer by Diffusion, Broken by Context: Diffusion LLM's Safety Blessing and Its Failure Modecited by 4
  8. 2025Adversarial-Guided Diffusion for Multimodal LLM Attackscited by 3
  9. 2025Falcon: A Cross-Modal Evaluation Dataset for Comprehensive Safety Perceptioncited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026GPO-V: Jailbreak Diffusion Vision Language Model by Global Probability Optimization
  2. 2025Diffusion-based Cumulative Adversarial Purification for Vision Language Models
  3. 2024Failures to Find Transferable Image Jailbreaks Between Vision-Language Models
  4. 2025VERA-V: Variational Inference Framework for Jailbreaking Vision-Language Models
  5. 2026Toward Universal and Transferable Jailbreak Attacks on Vision-Language Models
  6. 2026Mosaic: Multimodal Jailbreak against Closed-Source VLMs via Multi-View Ensemble Optimization
  7. 2025Safeguarding Vision-Language Models: Mitigating Vulnerabilities to Gaussian Noise in Perturbation-based Attacks
  8. 2025Dynamic Token Reweighting for Robust Vision-Language Models