ChatGPT Health Undertriages Serious Medical Emergencies

Researchers at the Icahn School of Medicine at Mount Sinai report in Nature Medicine (Feb 23, 2026) that ChatGPT Health under-triaged more than half of physician-classified emergency cases and produced inconsistent suicide-crisis alerts. The team tested 60 clinical scenarios across 21 specialties with 960 interactions, finding failures in nuanced emergencies and missed high-risk self-harm plans, and urges direct care seeking and ongoing independent evaluation.