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

2026-09-28 · ArXiv

arXiv:2609.26174v2 Announce Type: replace-cross

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 (

原文链接：https://arxiv.org/abs/2609.26174

查看更多资讯请见 AI 行业资讯。