Experts sound alarm after ChatGPT Health fails to recognise medical emergencies

4 hours ago

ChatGPT Health frequently misses the need for urgent medical care and fails to detect suicidal ideation, posing risks of harm or death.
A study found ChatGPT Health under-triaged over 50% of urgent cases, advising patients to stay home or book routine appointments instead of seeking emergency care.
In simulations, ChatGPT Health often downplayed severe symptoms, such as respiratory failure or diabetic ketoacidosis, creating a dangerous false sense of security.
The AI was inconsistent in responding to suicidal ideation, sometimes failing to display crisis intervention banners when lab results were mentioned.
Experts warn that ChatGPT Health could lead to unnecessary harm, legal liability, and inadequate emergency responses without stronger safeguards and oversight.
OpenAI claims the study doesn't reflect real-world usage and emphasizes continuous updates, but researchers urge clearer safety standards and independent audits.

Hasty Briefsbeta