Skip to content

The Uncontrolled Variable: Vision-Language Refusal Is Conditioned on the Image-Attachment Interface, and Not Robust to Irrelevant Image Properties

Aug 2026 · 0 citations · 26 references
Computer Science

Abstract

We show that aligned vision-language models also condition refusal on a property of a request's form: whether an image is attached, holding everything the request asks fixed. Attaching a blank canvas, an image that cannot be read, cannot relate to the request, and is byte-identical across every prompt in its condition, shifts benign refusal by tens of points. The shift is not blanket caution but a threshold shift: genuinely neutral instructions are almost unaffected (<=2 percentage points on three of four hosted models) while borderline-benign prompts move +23 to +51 points, so the cost falls on sensitivity-adjacent traffic, meaning benign questions about privacy, self-harm, violence and illegal activity. There is a benign reading of such a threshold, namely that attachment correlates with risk in real traffic, and we take it seriously; a black-box study cannot measure that correlation and we do not claim to. What it can test is whether the response to attachment is robust to variation carrying no information about the request, and on four independent measurements it is not. It varies with canvas colour and pixel count. Its sign inverts across checkpoints. It survives an explicit instruction to disregard the image. And on one model it fires on a bare assertion that an attachment exists, with nothing attached and the modality word contributing none of it. Finally we price it. On a matched harmful set the same canvas does lower attack success, so the cue buys something. But the charge is decoupled from the purchase: the checkpoint with the least harmful headroom we measure, completing only 2% of plain harmful requests, still pays the benign cost in full, and across our models the harmful-side denominator falls as alignment improves while the benign cost does not track it down. Image presence is not a conservative default that a deployer chose and priced. It is an uncontrolled variable.

View source

Similar papers

#artificial intelligence Open access May 2023

Evaluating the Performance of Large Language Models on GAOKAO Benchmark

GAOKAO-Bench is introduced, an intuitive benchmark that employs questions from the Chinese GAOKAO examination as test samples, including both subjective and objective questions that contribute a robust evaluation benchmark for future large language models and offers valuable insights into the advantages and limitations...

Xiaotian Zhang, Chun-yan Li, Yi Zong et al. · 216 citations · ⚡17
#artificial intelligence Open access Jul 2024

Gender, Race, and Intersectional Bias in Resume Screening via Language Model Retrieval

This work investigates the possibilities of using LLMs in a resume screening setting via a document retrieval framework that simulates job candidate selection and finds that the MTEs are biased, significantly favoring White-associated names in 85% of cases and female-associated names in only 11.1% of cases.

Kyra Wilson, Aylin Caliskan · 131 citations · ⚡8
#artificial intelligence Review Oct 2025

Ultralytics YOLO Evolution: An Overview of YOLO26, YOLO11, YOLOv8 and YOLOv5 Object Detectors for Computer Vision and Pattern Recognition

This paper presents a comprehensive overview of the Ultralytics YOLO family, emphasizing architectural evolution, benchmarking, deployment, and emerging directions from YOLOv5 through YOLO27, and examines detection, segmentation, depth, classification, pose, oriented detection, tracking, export, quantization, and deplo...

Ranjan Sapkota, Manoj Karkee · 112 citations · ⚡10

BadRAG: Identifying Vulnerabilities in Retrieval Augmented Generation of Large Language Models

A novel threat is unveiled in which attackers steer the RAG system's response by injecting malicious passages into its knowledge base, enabling the attacker to steer the response without altering the user input or modifying the RAG weights.

Jiaqi Xue, Meng Zheng, Yebowen Hu et al. · 109 citations · ⚡8

The Death of Schema Linking? Text-to-SQL in the Age of Well-Reasoned Language Models

This work revisits schema linking when using the latest generation of large language models (LLMs) and finds empirically that newer models are adept at utilizing relevant schema elements during generation even in the presence of large numbers of irrelevant ones.

Karime Maamari, Fadhil Abubaker, Daniel Jaroslawicz et al. · 109 citations · ⚡19

OverThink: Slowdown Attacks on Reasoning LLMs

This work evaluates Overthink on proprietary and open-source reasoning models across the FreshQA, SQuAD, and MuSR datasets, and shows that newer generations of RLMs, while showing a drastic increase in per-token cost, also exhibit up to a 2.3x increase in reasoning tokens, leaving them more vulnerable to Overthink atta...

Abhinav Kumar, Jaechul Roh, Ali Naseh et al. · 92 citations · ⚡9

Related blog posts

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.