Doctors are turning to AI to draft clinical notes, analyze data and make diagnoses, promising faster care and improved outcomes. However, because these systems learn from human records, they often inherit biases when factors such as gender or ethnicity are swapped.