Study Warns: ChatGPT Does Not Always Recognize Medical Emergencies

New research shows that AI tools like ChatGPT may underestimate situations requiring emergency medical attention, despite millions of people using them for health advice. This was reported by Euronews.

According to OpenAI data, more than 230 million people per week ask health-related questions. People ask everything from whether a certain food is safe to consume to managing allergies or treating a cold.

According to the new study published in Nature, the tool performs well with clear and obvious medical emergencies. In a number of cases, however, it fails to recommend seeking medical help in a timely manner.

The study, conducted by researchers from the Mount Sinai Health System, analyzed 60 clinical scenarios across 21 medical specialties—ranging from mild symptoms suitable for home treatment to life-threatening conditions. For each scenario, three independent physicians determined the correct level of urgency based on guidelines from 56 medical organizations.

The results show that the system correctly identifies some typical emergency conditions, such as a stroke or a severe allergic reaction. However, in more than half of the cases that actually require emergency medical intervention, the tool does not recommend an immediate visit to an emergency department.

In one of the analyzed scenarios involving an asthma attack, the system even recognized the early signs of respiratory failure in its explanation. Nevertheless, it advised the patient to wait instead of seeking emergency help.

The study also examined the system’s response to reports of self-harm. Although the platform is programmed to encourage seeking help and provide links to crisis hotlines, the “Help is available” warning appeared inconsistently.

The ChatGPT Health feature was introduced in January 2026. It allows users to connect health data—such as medical records or information from apps like MyFitnessPal—to receive more personalized responses.

The authors emphasize that the results do not mean that AI health tools should be abandoned. Rather, they should be used cautiously and should not replace medical expertise.

According to the researchers, individuals with symptoms such as chest pain, shortness of breath, severe allergic reactions, or sudden changes in consciousness should seek medical attention immediately. They should not rely solely on advice from a chatbot.

They further note that artificial intelligence systems are constantly being updated. This means that their results and accuracy can change over time and require continuous monitoring and evaluation.