AI Evaluations of Job Candidates Look Unbiased Without Being Fair
A Yale study of 26,000 GPT evaluations found that the model rarely ranked women or Black founders last but never made them more likely to win, and corrected openly biased human judgements far more than bias disguised as business sense. We explain both experiments, what symbolic compliance is, how it fits other AI hiring bias research, why a median-based bias audit can miss it, and the tests employers and investors should run.
