mlsanityData to 7 Oct 2026

StructBreak: Structural Cognitive Overload-Induced Safety Failures in MLLMs

arXiv 2605.25534 ↗First posted 25 May 2026cs.AIcited by 0

On the map: LLM reasoning → Reasoning for agents, planning and code

also in 1 more
Abstract (arXiv)

Multimodal Large Language Models (MLLMs) excel at structural reasoning yet suffer from a sharp logical brittleness in structural consistency. We term this phenomenon Structural Cognitive Overload (SCO), a byproduct of the contention between deep reasoning and safety alignment. However, prior work has predominantly targeted typographic and pixel-level perturbations, leaving the study of SCO largely unexplored. To this end, we propose StructBreak, an automated end-to-end framework designed to quantify SCO. By leveraging StructBreak, we uncover a novel higher-order cognitive overload attack paradigm; notably, this attack operates under a practical black-box setting, requiring no internal model access. Consequently, we utilize this framework to establish a comprehensive benchmark spanning ten diverse threat scenarios. Empirical evaluations on six leading MLLMs reveal that SCO readily triggers toxic generation, yielding a 92% average ASR (up to 97% on Gemini 2.5). To elucidate the mechanism of SCO, we further conduct model-level interpretations spanning attention dynamics, latent space topology, and geometric analysis. Our findings reveal that StructBreak acts as a novel structural channel to circumvent safety filters. Furthermore, the limited efficacy of inherent safety mechanisms underscores that current alignment paradigms are insufficient for the era of complex multimodal reasoning.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 25 of its 38 references that are in the map. First posted on arXiv 25 May 2026.

Builds on

25 of 38 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025Chain-of-Thought Hijacking30 Oct
  2. 2025A Mousetrap: Fooling Large Reasoning Models for Jailbreak with Chain of Iterative Chaos19 Feb
  3. 2024Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation14 Mar
  4. 2024How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMs12 Jan
  5. 2023Cognitive Overload: Jailbreaking Large Language Models with Overloaded Logical Thinking16 Nov

Other references · 20

  1. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  2. 2023Improved Baselines with Visual Instruction Tuningcited by 4,362
  3. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Promptscited by 368
  4. 2023Hard Prompts Made Easy: Gradient-Based Discrete Optimization for Prompt Tuning and Discoverycited by 329
  5. 2023Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Modelscited by 260
  6. 2021Gradient-based Adversarial Attacks against Text Transformerscited by 233
  7. 2024ArtPrompt: ASCII Art-based Jailbreak Attacks against Aligned LLMscited by 197
  8. 2024Jailbreak Vision Language Models via Bi-Modal Adversarial Promptcited by 103
  9. 2024ImgTrojan: Jailbreaking Vision-Language Models with ONE Imagecited by 50
  10. 2024ChartMoE: Mixture of Diversely Aligned Expert Connector for Chart Understandingcited by 44
  11. 2025Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategycited by 37
  12. 2025Control Illusion: The Failure of Instruction Hierarchies in Large Language Modelscited by 31
8 more
  1. 2025HiddenDetect: Detecting Jailbreak Attacks against Large Vision-Language Models via Monitoring Hidden Statescited by 29
  2. 2024FlowVQA: Mapping Multimodal Logic in Visual Question Answering with Flowchartscited by 27
  3. 2025FC-Attack: Jailbreaking Multimodal Large Language Models via Auto-Generated Flowchartscited by 15
  4. 2024Gradient-based Jailbreak Images for Multimodal Fusion Modelscited by 13
  5. 2024Cognitive Overload Attack:Prompt Injection for Long Contextcited by 13
  6. 2025LLMs can be easily Confused by Instructional Distractionscited by 12
  7. 2025MAPS: Advancing Multi-Modal Reasoning in Expert-Level Physical Sciencecited by 10
  8. 2025SEA: Low-Resource Safety Alignment for Multimodal Large Language Models via Synthetic Embeddingscited by 9

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025SafeMLRM: Demystifying Safety in Multi-modal Large Reasoning Models
  2. 2025When Safe Unimodal Inputs Collide: Optimizing Reasoning Chains for Cross-Modal Safety in Multimodal Large Language Models
  3. 2025SafeRBench: Dissecting the Reasoning Safety of Large Language Models
  4. 2025Think in Safety: Unveiling and Mitigating Safety Alignment Collapse in Multimodal Large Reasoning Model
  5. 2026ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models
  6. 2025VisCRA: A Visual Chain Reasoning Attack for Jailbreaking Multimodal Large Language Models
  7. 2025SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment
  8. 2026MME-Safety: A Fine-grained Benchmark for Safety Evaluation of MLLMs