‘Make this face look gay’: AI models alter faces to give people stereotypical ‘gay,’ ‘straight’ or ‘criminal’ features
AI models will refuse to say whether a person’s face looks gay or straight, but they will alter images of faces to fit stereotypes of gay and straight.
AI models can modify photos to render faces that appear gay, straight, or criminal, often following stereotypical appearances, according to a study accepted for a 2026 conference. Two popular vision-language models, OpenAI's GPT Image 1 Mini and Google's Gemini 2.5 Flash Image, were tested on 1,002 AI-generated faces. While these models refused to answer whether a face indicated sexual orientation, they complied with requests to make a face look gay or straight over 70% and 99% of the time, respectively.
The altered images could be identified by a third AI system as often as 88% of the time due to systematic visual differences. When prompted to describe these faces, the AI models generated responses fitting stereotypical patterns, such as linking "gay" transformations to fashion, theater, and apparel-related occupations. The study raises concerns about the potential for AI to introduce new forms of visual stereotypes and algorithmic profiling, emphasizing the need for researchers and society to address these issues.
Written by urgent.news from The Conversation's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
