HARVARD SAYS AI BEATS YOUR DOCTOR — AND IT IS NOT EVEN CLOSE A study out of Harvard Medical School and Beth Israel Deaconess Medical Center has confirmed what the AI labs have been claiming for years. An OpenAI reasoning model outperformed human emergency room physicians on diagnosis with a consistency that the lead researchers called alarming. The AI offered exact or very close diagnoses in 67 percent of initial triage cases. Human doctors hit 50 to 55 percent under identical conditions with the same information. It gets worse for the physicians when the cases run longer. On five extended clinical scenarios where doctors were allowed to use search engines and any other resources they wanted, the AI scored 89 percent. The 46 human doctors averaged 34 percent. That is not a close race. That is not a promising early result from a new technology. That is a systematic, repeatable gap that shows up across different kinds of cases, different conditions, and different stages of care. The researchers were careful not to call for replacing doctors. They pointed to real limitations: text-only inputs, no physical examination, no imaging, no patient interaction. Real clinical medicine is messier than structured case files. All of that is true. None of it changes what the numbers say. AI is now performing at or above the level of trained specialists in one of the most consequential domains that exists. The study published April 30. The 76 cases came from real Beth Israel ER admissions with no cleanup before being fed to the model. The implications are not going away. Keywords: Harvard AI medical study, AI beats doctors diagnosis, OpenAI clinical medicine, AI emergency room triage