Architecture Matters: Gender Disparities in Automated Image Moderation

Autorzy Zawadzka A.; Głomb P.
Tytuł Architecture Matters: Gender Disparities in Automated Image Moderation
Czasopismo PP-RAI'2026: 7th Polish Conference on Artificial Intelligence
Rok 2026
Status Published
Abstrakt <p>Automated image moderation systems shape online visibility and dataset curation, yet prior work has identified demographic disparities in similarity-based NSFW classifiers. We compare such systems with instruction-tuned vision-language models (VLMs) that generate structured moderation decisions with textual rationales. Using the PHASE-annotated subset of the GCC dataset, &nbsp;we compute false positive rates overall and by gender.&nbsp;<br>Results show substantial variation in moderation strictness and in bias direction: two models show more pronounced strictness in moderation of male images, while other two exhibit higher removal rates for female images.&nbsp;<br>Reasoning-based models do not consistently mitigate bias but, in the proposed pipeline, they offer a transparency layer.</p>