mlsanityData to 14 Oct 2026

Hard to Read, Easy to Jailbreak: How Visual Degradation Bypasses MLLM Safety Alignment

arXiv 2605.07250 ↗First posted 8 May 2026cs.CV cs.AIcited by 1

On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks

also in 2 more
Abstract (arXiv)

Recent advancements in visual context compression enable MLLMs to process ultra-long contexts efficiently by rendering text into images. However, we identify a critical vulnerability inherent to this paradigm: lowering image resolution inadvertently catalyzes jailbreaking. Our experiments reveal that the safety defenses of SOTA models deteriorate sharply as resolution degrades, surprisingly persisting even when text remains legible. We attribute this to ``Cognitive Overload'', hypothesizing that the effort required to decipher degraded inputs diverts attentional resources from safety auditing. This phenomenon is consistent across various visual perturbations, including noise and geometric distortion. To address this, we propose a simple ``Structured Cognitive Offloading'' strategy that mitigates these risks by enforcing a serialized pipeline to decouple visual transcription from safety assessment. Our work exposes a significant risk in vision-based compression and provides critical insights for the secure design of future MLLMs.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 16 of its 26 references that are in the map. First posted on arXiv 8 May 2026.

Builds on

16 of 26 references.

Closest ancestors · 2

Same direction, newest first

  1. 2024Safety Layers in Aligned Large Language Models: The Key to LLM Security30 Aug
  2. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts9 Nov

Other references · 14

  1. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,796
  2. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
  3. 2025GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learningcited by 466
  4. 2023Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Modelscited by 273
  5. 2024JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attackscited by 230
  6. 2023How Robust is Google's Bard to Adversarial Image Attacks?cited by 183
  7. 2023Image Hijacks: Adversarial Images can Control Generative Models at Runtimecited by 172
  8. 2024Making Them Ask and Answer: Jailbreaking Large Language Models in Few Queries via Disguise and Reconstructioncited by 107
  9. 2023On the Robustness of Large Multimodal Models Against Image Adversarial Attackscited by 92
  10. 2024Defending Large Language Models Against Jailbreak Attacks via Layer-specific Editingcited by 79
  11. 2025Glyph: Scaling Context Windows via Visual-Text Compressioncited by 49
  12. 2025JPS: Jailbreak Multimodal Large Language Models with Collaborative Visual Perturbation and Textual Steeringcited by 11
2 more
  1. 2024Jailbreaking? One Step Is Enough!cited by 9
  2. 2025Demystifying the Visual Quality Paradox in Multimodal Large Language Modelscited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026PHANTOM: A Large-Scale Dataset of Multimodal Adversarial Attacks for Vision-Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Beyond Visual Safety: Jailbreaking Multimodal Large Language Models for Harmful Image Generation via Semantic-Agnostic Inputs
  2. 2026When Think-with-Image Meets Safety: What Determines Multimodal Jailbreak Robustness?
  3. 2026Scenes as an Attack Surface: Uncovering Scene-Grounded Safety Failures in MLLMs
  4. 2026Visual Token Compression Enhances Robustness of MLLMs
  5. 2026Less Is More -- Until It Breaks: Security Pitfalls of Vision Token Compression in Large Vision-Language Models
  6. 2024Learning To See But Forgetting To Follow: Visual Instruction Tuning Makes LLMs More Prone To Jailbreak Attacks
  7. 2026One Perturbation, Two Failure Modes: Probing VLM Safety via Embedding-Guided Typographic Perturbations
  8. 2026COMIC: Reference-Aware Safety Gating for Multimodal Large Language Models