Study Finds ChatGPT Health Underestimates Urgency in Over Half of Emergency Cases

A study published in Nature Medicine evaluated OpenAI's ChatGPT Health chatbot's ability to triage medical cases, finding it under-triaged about 51-52% of emergency scenarios. Researchers from Mount Sinai presented 60 simulated cases, comparing the chatbot's recommendations with those of physicians. The AI often delayed urgent care advice, suggesting non-emergency visits for serious conditions like diabetic ketoacidosis and respiratory failure. While it identified clear emergencies, the tool struggled with complex cases, prompting calls for further testing before clinical use.