Story perspectives

Study Reveals ChatGPT Health Fails 52% of Medical Emergencies

2/26/2026

Summary

Story summary

A Nature Medicine study evaluated ChatGPT Health, OpenAI's health guidance tool launched in January 2026, and found it under-triaged over 52% of medical emergencies.
In critical scenarios like respiratory failure, the AI waited instead of prompting immediate medical care.
The system inconsistently activated suicide-risk alerts and failed to respond to users with explicit self-harm plans.
Researchers urge rigorous independent evaluations to ensure safety and reliability of AI tools in healthcare.

1 of 1

Story summary

A Nature Medicine study evaluated ChatGPT Health, OpenAI's health guidance tool launched in January 2026, and found it under-triaged over 52% of medical emergencies.
In critical scenarios like respiratory failure, the AI waited instead of prompting immediate medical care.
The system inconsistently activated suicide-risk alerts and failed to respond to users with explicit self-harm plans.
Researchers urge rigorous independent evaluations to ensure safety and reliability of AI tools in healthcare.