Artificial intelligence safety filters are almost completely eliminating female characters from children's fairy tales, reports infohub.kz.
Researchers from the University of Washington's Information School, led by Imani Finkley and Melanie Walsh, analyzed nearly 24,000 AI-generated variations of children's animal stories. The experiment tested leading AI models, including OpenAI's GPT-5.1, Google's Gemini 2.5, Anthropic's Claude Sonnet 4.5, and the open-source model Olmo 3.
The results showed that algorithms designed to prevent gender bias had the opposite effect. In 57% of cases, animals were assigned the neuter pronoun "it," 41% of characters were male, and only 2% were female. Gemini 2.5 and GPT-5.1 were the most likely to create male characters (63% and 65% of stories, respectively).
The researchers used this experiment as a kind of "Bechdel test" for algorithms. According to the authors, the developers' attempt to achieve neutrality did not correct bias but instead led to the effective erasure of female characters from children's stories.