Artificial intelligence is increasingly used to answer medical questions, but new research suggests it may still struggle with high-stakes decisions. A recent study tested a health-focused AI chatbot designed to help users assess symptoms and found it frequently underestimated the severity of medical emergencies.
Researchers evaluated the chatbot using 60 real-world medical scenarios and compared its responses with those of physicians reviewing the same cases. The AI under-triaged more than half of emergency situations, often advising users to see a doctor within a day or two instead of seeking immediate emergency care.
These included life-threatening conditions such as severe diabetes complications and impending respiratory failure. At the same time, the chatbot sometimes overreacted to minor issues, recommending medical visits for symptoms that could typically be managed at home.
The findings highlight a broader concern about relying on AI for urgent health decisions. While chatbots can provide quick answers and are increasingly used outside normal clinic hours, researchers say performance in controlled tests does not necessarily translate to safe real-world use.
The study also found inconsistent responses in scenarios involving suicidal thoughts. Experts say AI tools may still play a role in health information and patient education, but they should not replace clinical judgment, especially in emergencies.
SOURCE:
NBC News, March 3, 2026