mlsanityData to 7 Oct 2026

Safety-Potential Pruning for Enhancing Safety Prompts Against VLM Jailbreaking Without Retraining

arXiv 2603.14219 ↗First posted 15 Mar 2026cs.CVcited by 0

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

also in 1 more
Abstract (arXiv)

Safety prompts constitute an interpretable layer of defense against jailbreak attacks in vision-language models (VLMs); however, their efficacy is constrained by the models' latent structural responsiveness. We observe that such prompts consistently engage a sparse set of parameters that remain largely quiescent during benign use. This finding motivates the Safety Subnetwork Hypothesis: VLMs embed structurally distinct pathways capable of enforcing safety, but these pathways remain dormant without explicit stimulation. To expose and amplify these pathways, we introduce Safety-Potential Pruning, a one-shot pruning framework that amplifies safety-relevant activations by removing weights that are less responsive to safety prompts without additional retraining. Across three representative VLM architectures and three jailbreak benchmarks, our method reduces attack success rates by up to 22% relative to prompting alone, all while maintaining strong benign performance. These findings frame pruning not only as a model compression technique, but as a structural intervention to emerge alignment-relevant subnets, offering a new path to robust jailbreak resistance.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 49 of its 57 references that are in the map. First posted on arXiv 15 Mar 2026.

Builds on

49 of 57 references.

Closest ancestors · 16

Same direction, newest first

  1. 2025Hierarchical Safety Realignment: Lightweight Restoration of Safety in Pruned Large Vision-Language Models22 May
  2. 2025Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models30 Jan
  3. 2024Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks23 Nov
  4. 2024Cross-Modal Safety Mechanism Transfer in Large Vision-Language Models16 Oct
  5. 2024Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Models11 Oct
  6. 2024ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time9 Oct
  7. 2024CoCA: Regaining Safety-awareness of Multimodal Large Language Models with Constitutional Calibration17 Sep
  8. 2024JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Models26 Jun
  9. 2024JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks3 Apr
  10. 2024Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation14 Mar
  11. 2024AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting14 Mar
  12. 2024MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance5 Jan
4 more
  1. 2023MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models29 Nov
  2. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts9 Nov
  3. 2023Can Language Models be Instructed to Protect Personal Information?3 Oct
  4. 2023On Evaluating Adversarial Robustness of Large Vision-Language Models26 May

Other references · 33

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  3. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  4. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  5. 2023Visual Instruction Tuningcited by 7,792
  6. 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,069
  7. 2018GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understandingcited by 4,854
  8. 2022BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generationcited by 4,310
  9. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,151
  10. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 2,914
  11. 2023InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuningcited by 2,855
  12. 2018Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answeringcited by 2,114
21 more
  1. 2023MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGIcited by 2,044
  2. 2015Learning both Weights and Connections for Efficient Neural Networkscited by 1,998
  3. 2022Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answeringcited by 1,881
  4. 2019Towards VQA Models That Can Readcited by 1,696
  5. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,696
  6. 2019VisualBERT: A Simple and Performant Baseline for Vision and Languagecited by 1,291
  7. 2023MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilitiescited by 1,221
  8. 2023SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shotcited by 1,090
  9. 2023A Simple and Effective Pruning Approach for Large Language Modelscited by 776
  10. 2024VLMEvalKit: An Open-Source Toolkit for Evaluating Large Multi-Modality Modelscited by 443
  11. 2021VLMo: Unified Vision-Language Pre-Training with Mixture-of-Modality-Expertscited by 424
  12. 2023Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignmentcited by 394
  13. 2024Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modificationscited by 220
  14. 2023Survey of Vulnerabilities in Large Language Models Revealed by Adversarial Attackscited by 189
  15. 2024Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Modelscited by 160
  16. 2024On Prompt-Driven Safeguarding for Large Language Modelscited by 109
  17. 2023DRESS: Instructing Large Vision-Language Models to Align and Interact with Humans via Natural Language Feedbackcited by 99
  18. 2018Sparse DNNs with Improved Adversarial Robustnesscited by 71
  19. 2024Pruning for Protection: Increasing Jailbreak Resistance in Aligned LLMs Without Fine-Tuningcited by 24
  20. 2024The First to Know: How Token Distributions Reveal Hidden Knowledge in Large Vision-Language Models?cited by 24
  21. 2019Towards Compact and Robust Deep Neural Networkscited by 23

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Understanding and Mitigating Token-Pruning-Induced Vulnerabilities in VLMs
  2. 2026Understanding and Defending VLM Jailbreaks via Jailbreak-Related Representation Shift
  3. 2025Defense-to-Attack: Bypassing Weak Defenses Enables Stronger Jailbreaks in Vision-Language Models
  4. 2025HiddenDetect: Detecting Jailbreak Attacks against Large Vision-Language Models via Monitoring Hidden States
  5. 2026Circuit Discovery Helps Detect LLM Jailbreaking: A Mechanistic Interpretability Study
  6. 2025JailBound: Jailbreaking Internal Safety Boundaries of Vision-Language Models
  7. 2025Beyond Surface Alignment: Rebuilding LLMs Safety Mechanism via Probabilistically Ablating Refusal Direction
  8. 2025How Jailbreak Defenses Work and Ensemble? A Mechanistic Investigation