Advanced AI Models Continue to Reproduce Racial and Gender Stereotypes in Medical Contexts

Flinders University researchers have found that next-generation reasoning Large Language Models (LLMs), including o3-mini and DeepSeek-R1, frequently reproduce racial and gender stereotypes when generating descriptions of fictional patients with common medical conditions. Published in the Journal of Medical Internet Research, this finding indicates that advancements in AI reasoning capabilities do not inherently improve representational fairness, highlighting a critical challenge for the safe and equitable integration of AI into healthcare and scientific discovery.