AI Outperforms ER Doctors in Real Cases

A study from Harvard Medical School and Beth Israel Deaconess shows an OpenAI reasoning model achieved higher diagnostic and management accuracy than two experienced emergency physicians when given only the electronic health record text at the same timepoints doctors had it. The result held across messy, retrospective emergency‑department records and established clinical vignettes, though authors caution the system lacked non‑text inputs and call for prospective trials before clinical deployment. — If validated prospectively, this finding could reshape emergency‑department workflows, liability rules, staffing models, and regulatory standards for clinical AI tools.