Vision-Language Models Suppress Female Representations Under Ambiguous Input

ArXi:2605.31556v1 Announce Type: cross Alignment teaches vision-language models (VLMs) to avoid expressing graphic biases, and when gender is clearly visible they largely succeed. Far less is known about ambiguous inputs (a worker in full gear, a figure seen from behind) cases common in practice yet rarely studied. We find that minimal prompting pressure exposes occupation-gender defaults when prompting ambiguous input images, with models collapsing to male even for strongly female-stereotyped occupations. But do these outputs reflect what models actually encode internally? We.