A study published in Nature reveals that while ChatGPT Health performs well in textbook medical emergencies, it often underestimates the urgency of serious cases, advising users to wait instead of seeking emergency care. The study, led by Ashwin Ramaswamy at Mount Sinai, found inconsistencies in the model's responses, particularly regarding self-harm intentions. Despite these issues, researchers advocate for the thoughtful integration of AI health tools into medical care rather than abandoning them.