mlsanityData to 7 Oct 2026

Still There, No Longer Seen: Exposing Compression-Induced Risk in Large Vision-Language Models

arXiv 2609.35002 ↗First posted 28 Sep 2026cs.CV cs.AI cs.CRcited by 0

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness also fits: Visual token pruning and efficiency

also in 1 more
Abstract (arXiv)

Visual token compression reduces the inference cost of Large Vision-Language Models (LVLMs). However, aggregate robustness measures do not reveal whether a particular adversarial failure is induced by compression or inherited from the underlying model. We define a compression-specific failure (CSF) as an adversarial input that remains correct under full-token inference but fails after compression, casting compression-induced risk as a paired failure attribution problem. Within a controlled diagnostic cohort, counterfactuals show that retained-set allocation causally changes compressed correctness and reveal a negative association between recovery and representation drift in displaced evidence. Motivated by these findings, we propose CIRA, a Compression-Induced Risk Attack for Large Vision-Language Models. Under a vision-encoder white-box setting, CIRA optimizes image perturbations through encoder-side objectives that manipulate token priorities across candidate compression budgets while preserving displaced evidence. CIRA uses no downstream questions or labels and requires no access to the language model, deployed compressor, or exact compression budget. Across 12 dataset-compressor settings evaluated at four budgets, CIRA achieves a mean CSFR of 20.35% while limiting full-token attack success to 6.92%, with similar behavior on additional LVLM families. A cross-view selection-stabilization defense substantially suppresses CIRA, although Adaptive CIRA partially restores its effectiveness. These results show that compression-specific failures persist under restricted access and support paired evaluation of full-token and compressed inference for attributing risk to visual-token compression.

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 51 of its 58 references that are in the map. First posted on arXiv 28 Sep 2026.

Builds on

51 of 58 references.

Closest ancestors · 18

Same direction, newest first

  1. 2026Visual Token Compression Enhances Robustness of MLLMs21 Jul
  2. 2026PA-Attack: Guiding Gray-Box Attacks on LVLM Vision Encoders with Prototypes and Attention23 Feb
  3. 2026Grounding-Driven Attack: Improving Encoder-based Adversarial Transferability against Large Vision-Language Models10 Feb
  4. 2026On the Adversarial Robustness of Large Vision-Language Models under Visual Token Compression29 Jan
  5. 2026Less Is More -- Until It Breaks: Security Pitfalls of Vision Token Compression in Large Vision-Language Models17 Jan
  6. 2025Self-Calibrated Consistency can Fight Back for Adversarial Robustness in Vision-Language Models26 Oct
  7. 2025VEAttack: Downstream-agnostic Vision Encoder Attack against Large Vision Language Models23 May
  8. 2025Transferable Adversarial Attacks on Black-Box Vision-Language Models2 May
  9. 2025R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning15 Apr
  10. 2025Double Visual Defense: Adversarial Pre-training and Instruction Tuning for Improving Vision-Language Model Robustness16 Jan
  11. 2024Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks24 Nov
  12. 2024AnyAttack: Towards Large-scale Self-supervised Adversarial Attacks on Vision-language Models7 Oct
6 more
  1. 2024An Image Is Worth 1000 Lies: Adversarial Transferability across Prompts on Vision-Language Models14 Mar
  2. 2024Robust CLIP: Unsupervised Adversarial Fine-Tuning of Vision Embeddings for Robust Large Vision-Language Models19 Feb
  3. 2023Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models26 Jul
  4. 2023Set-level Guidance Attack: Boosting Adversarial Transferability of Vision-Language Pre-training Models26 Jul
  5. 2023Visual Adversarial Examples Jailbreak Aligned Large Language Models22 Jun
  6. 2022Towards Adversarial Attack on Vision-Language Pre-training Models19 Jun

Other references · 33

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,069
  3. 2023Improved Baselines with Visual Instruction Tuningcited by 4,362
  4. 2022Flamingo: a Visual Language Model for Few-Shot Learningcited by 4,337
  5. 2025Qwen3-VL Technical Reportcited by 2,534
  6. 2019Towards VQA Models That Can Readcited by 1,696
  7. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,696
  8. 2023MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Modelscited by 1,624
  9. 2024An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Modelscited by 629
  10. 2024LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Modelscited by 394
  11. 2024SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inferencecited by 378
  12. 2024LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understandingcited by 341
21 more
  1. 2024VisionZip: Longer is Better but Not Necessary in Vision Language Modelscited by 298
  2. 2024PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reductioncited by 283
  3. 2025DivPrune: Diversity-based Visual Token Pruning for Large Multimodal Modelscited by 177
  4. 2024TokenPacker: Efficient Visual Projector for Multimodal LLMcited by 170
  5. 2024Beyond Text-Visual Attention: Exploiting Visual Cues for Effective Token Pruning in VLMscited by 153
  6. 2023On the Adversarial Robustness of Multi-Modal Foundation Modelscited by 142
  7. 2024ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Modelscited by 89
  8. 2023VLATTACK: Multimodal Adversarial Attacks on Vision-Language Tasks via Pre-trained Modelscited by 66
  9. 2025PACT: Pruning and Clustering-Based Token Reduction for Faster Visual Language Modelscited by 49
  10. 2024Feather the Throttle: Revisiting Visual Token Pruning for Vision-Language Model Accelerationcited by 40
  11. 2024Dynamic-VLM: Simple Dynamic Visual Token Compression for VideoLLMcited by 31
  12. 2025Variation-aware Vision Token Dropping for Faster Large Vision-Language Modelscited by 25
  13. 2026VisionTrim: Unified Vision Token Compression for Training-Free MLLM Accelerationcited by 16
  14. 2026N"uwa: Mending the Spatial Integrity Torn by VLM Token Pruningcited by 13
  15. 2025When Token Pruning is Worse than Random: Understanding Visual Token Information in VLLMscited by 11
  16. 2026LearnPruner: Rethinking Attention-based Token Pruning in Vision Language Modelscited by 7
  17. 2026Improving Visual Token Reduction via Rectifying Distortions for Efficient Multimodal LLM Inferencecited by 5
  18. 2025TransPrune: Token Transition Pruning for Efficient Large Vision-Language Modelcited by 3
  19. 2025Does Acceleration Cause Hidden Instability in Vision Language Models? Uncovering Instance-Level Divergence Through a Large-Scale Empirical Studycited by 2
  20. 2026Rethinking Token Reduction for Large Vision-Language Modelscited by 2
  21. 2026Reroute, Don't Remove: Recoverable Visual Token Routing for Vision-Language Modelscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Feature-Aware Token Attack for Compression-Triggered Stealthy Failures in Large Vision-Language Models
  2. 2025High-Entropy Tokens as Multimodal Failure Points in Vision-Language Models
  3. 2026Corrupted but Correct: Why Vision-Language Models Lie to Themselves Internally
  4. 2026Mechanistically Interpreting Compression in Vision-Language Models
  5. 2026Edge Reliability Gap in Vision-Language Models: Quantifying Failure Modes of Compressed VLMs Under Visual Corruption
  6. 2025Are We Using the Right Benchmark: An Evaluation Framework for Visual Token Compression Methods
  7. 2025LLMC+: Benchmarking Vision-Language Model Compression with a Plug-and-play Toolkit
  8. 2026VisCo: Leveraging Large Language Models as Intrinsic Encoders for Visual Token Compression