AI Develops New Hiring Biases From Experience

Researchers from Princeton University and the University of Chicago found that large language models, including ChatGPT, Claude, Gemini, OpenAI's o3, and DeepSeek's R1, spontaneously develop novel social biases from experience in a simulated hiring game. The models stereotyped candidates from fictional ethnic groups more severely than human participants, with o3 scoring 1.83 on a segregation scale where humans scored 0.84.