AI fails to diagnose skin cancer in patients with darker skin tones
Claude and ChatGPT artificial intelligence models showed limited accuracy in identifying skin cancers in patients with dark skin tones, despite expressing relatively high confidence in their diagnoses, a study found. The researchers, who presented their findings at the American Society for Dermatologic Surgery Annual Meeting in San Diego, reported the models diagnosed 13 of 48 darker skin cancers as benign with at least 70% confidence. They asked the chatbots to rate their diagnostic confidence ranging from 1% to 100%. For lesions in the darkest skin colors, the models reported confidence levels of 65% to 70% when their diagnoses were correct and 60% to 65% when they were wrong. The findings underscore that general-purpose AI should not be used for skin cancer triage. (Healio.com article, 11/10/26)