CORTEXA
← Browse
arxivcs.CVcs.LG2026-06-30

Steal the Patch Size: Adversarially Manipulate Vision-Language Models

Kai Hu, Akash Bharadwaj, Weichen Yu, Matt Fredrikson

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. The key idea is a task-level side channel induced by ViT-style patchification: when a synthetic grid image is aligned with the hidden patch grid, boundary cues are erased at tokenization, causing periodic accuracy drop. By sweeping the grid cell size and measuring these collapses, we infer the patch size; by introducing padding and a consistency-check test, we further identify whether preprocessing is dynamic- or fixed-resolution and recover the target resize resolution. Across open-source Qwen-VL variants and proprietary models including GPT and Claude, we reliably recover tokenizer-related parameters. Finally, we show that such leakage enables preprocessing-aware transfer attacks and model-targeted adversarial manipulation.

View free PDFSource page

Related papers

arxivcs.CVcs.LG2026-07-06

Is the Geometry Doing the Work? An Operating-Point Audit of Hierarchy in Hyperbolic Vision-Language Models

Jaeyoung Kim, Eunseok Kim, Dongsuk Jang

Hyperbolic vision-language models are designed to encode abstraction geometrically: general concepts near the origin, specific ones farther out, and entailment cones representing directed order. We ask whether trained MERU, HyCoCLIP, and PHyCLIP models actually use these mechanis…

View free PDFSource page
arxivcs.CVcs.LG2026-07-03

WorldBagel: Uncovering the Power of Unified Multimodal Models for Vision-Language-Action-World Modeling

Zelin Zhao, Min Shi, Bo Yuan, Haotian Xue, Jialuo Li, Lama Moukheiber, et al.

World models aim to capture environment dynamics in ways that support perception, reasoning, and action, and have recently become a central direction in Vision-Language-Action-World (VLAW) modeling. Meanwhile, unified vision-language models have demonstrated strong multimodal gen…

View free PDFSource page
arxivcs.CVcs.AIcs.LGeess.IV2026-07-21

Attributes Should Come from Images, Not Class Names: Distribution-Conditioned Attribute Selection for Vision-Language Models

Gautam Rajendrakumar Gare, Jia Shi, Zhiqiu Lin, Deepak Pathak, John Galeotti, Deva Ramanan

A popular route to interpretable zero-shot classification asks a large language model (LLM) to describe each class name and prompts CLIP with the resulting descriptors. We show that these descriptors carry little visual evidence of their own: removing the class name from the prom…

View free PDFSource page
arxivcs.CVcs.AIcs.LGeess.IV2026-07-17

Ask Twice, Look Twice: Prompt Echoing Resolves the Question-First Paradox in Vision-Language Models

Rakshanda Hassan Abhinandan, John Galeotti, Deva Ramanan, Gautam Rajendrakumar Gare

Where should the question go in a vision-language model (VLM) prompt: before the image or after it? Intuition says before: knowing what is asked should tell the model where to look. Yet across visual question answering benchmarks, question-first prompting consistently underperfor…

View free PDFSource page
arxivcs.CVcs.LG2026-07-10

Robustifying Vision-Language Models via Test-Time Prompt Adaptation

Xingyu Zhu, Huanshen Wu, Shuo Wang, Beier Zhu, Jiannan Ge, Jiaheng Zhang, et al.

Pre-trained Vision-Language Models (VLMs) such as CLIP achieve strong zero-shot generalization, but their performance degrades sharply under adversarial perturbations. Existing test-time adaptation methods typically rely on sample-level confidence heuristics, overlooking the intr…

View free PDFSource page
arxivcs.CVcs.CLcs.LG2026-07-03

Pathways of Visual Information Flow in Vision-Language Models

Israfel Salazar, Stella Frank, Dan Oneata, Desmond Elliott, Constanza Fierro

We study how visual information is routed in vision-language models (VLMs). Using causal patching on controlled synthetic and natural datasets, we find that models rely on two distinct pathways to solve visual tasks: A direct pathway, where visual information is retained in image…

View free PDFSource page