Detect Before You Leap: Mirage Detection in Vision-Language Models

Sayeed Shafayet Chowdhury, Md. Shaown Miah, S. M. Taiabul Haque +1

Vision-language models (VLMs) can produce confident answers without relevant visual evidence, a failure mode known as mirage reasoning (Asadi et al., 2026). To that end, we study p…