arXiv Machine Learning July 2, 2026 By Kai Hu, Akash Bharadwaj, Weichen Yu, Matt Fredrikson

Steal the Patch Size: Adversarially Manipulate Vision-Language Models

Read the original on arXiv Machine Learning →

arXiv:2607. 00174v1 Announce Type: cross Abstract: We present a black-box model-stealing attack that recovers private vision-tokenizer configurations of deployed vision-language models (VLMs), including the visual patch size and input preprocessing pipeline.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

arXiv AI

Jun 11

Diffusion-based Cumulative Adversarial Purification for Vision Language Models

arXiv:2506. 03933v2 Announce Type: replace-cross Abstract: Vision Language Models (VLMs) have shown remarkable capabilities in multimodal understanding, yet their susceptibility to adversarial perturbations poses a significant threat to their reliability in real-world applications.

By Jia Fu, Yongtao Wu, Yihang Chen, Kunyu Peng, Xiao Zhang, Volkan Cevher, Sepideh Pashami, Anders Holst

llms rag diffusion multimodal safety

arXiv AI

Sep 1

Robustness of Vision Language Models Against Split-Image Harmful Input Attacks

arXiv:2602.08136v2 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) are now a core part of modern AI. Recent work proposed several visual jailbreak attacks using single/ holistic...

By Md Rafi Ur Rashid, MD Sadik Hossain Shanto, Vishnu Asutosh Dasu, Shagufta Mehnaz

llms reinforcement-learning fine-tuning efficiency multimodal benchmarks safety

arXiv Computer Vision

3d ago

Feature-Aware Token Attack for Compression-Triggered Stealthy Failures in Large Vision-Language Models

arXiv:2609.39134v1 Announce Type: new Abstract: Visual-token compression improves the efficiency of large vision-language models, but can expose failures that full-token evaluation misses. We study a...

By Shilinlu Yan, Bowen Chen, Yuechen Zhang, Zhenhong Zhou, Li Sun, Sen Su

llms multimodal safety

arXiv Computer Vision

4d ago

Selective Channel Restoration for Backdoored Vision-Language Models

arXiv:2609.37759v1 Announce Type: cross Abstract: Vision-language models (VLMs) exhibit strong multimodal capabilities but remain vulnerable to backdoors implanted through poisoned fine-tuning data....

By Shuming Liu, Zhifang Zhang, Suqin Yuan, Khin Mi Mi Aung, Zhuoyi Lin, Lei Feng

llms fine-tuning multimodal

arXiv Machine Learning

Jun 11

Right Predictions, Misleading Explanations: On the Vulnerability of Vision-Language Model Explanations

arXiv:2605. 16651v2 Announce Type: replace-cross Abstract: Explanation mechanisms are increasingly used to support transparency and trust in vision-language models (VLMs), particularly in settings where model decisions require human oversight.

By Narges Babadi, Hadis Karimipour

llms multimodal safety

arXiv AI

Aug 20

Breaking the weakest link to evade vision language models

The paper investigates how Vision Language Models (VLMs) can be fooled by small, human‑imperceptible changes to images. It introduces a gradient‑based attack that targets only the vision encoder, reducing computational cost while still effectively disrupting both untargeted and targeted multimodal alignment. Experiments on open‑source VLMs such as Qwen2.5‑VL, Granite‑Vision, FastVLM, and Phi‑3.5‑Vision demonstrate that these perturbations can significantly alter the models’ textual outputs.

By Ilan Zini, Boussad Addad, Katarzyna Kapusta

llms robotics multimodal safety

One email a morning, machine-written

Related stories

Diffusion-based Cumulative Adversarial Purification for Vision Language Models

Robustness of Vision Language Models Against Split-Image Harmful Input Attacks

Feature-Aware Token Attack for Compression-Triggered Stealthy Failures in Large Vision-Language Models

Selective Channel Restoration for Backdoored Vision-Language Models

Right Predictions, Misleading Explanations: On the Vulnerability of Vision-Language Model Explanations

Breaking the weakest link to evade vision language models