Newer AI Models Still Reproduce Racial and Gender Stereotypes in Medicine

Researchers at Flinders University evaluated newer large language models, including o3-mini and DeepSeek-R1, and found that they frequently misrepresented the distribution of race and gender in descriptions of patients with common medical conditions. The findings suggest that improvements in AI reasoning do not automatically translate into representational fairness, highlighting ongoing risks for biased clinical decision support and patient safety.