ChatGPT Health inconsistently triggered suicide-crisis alerts cases in which users described specific plans for self-harm, according to a study published Feb. 23 in Nature Medicine. The evaluation — led by researchers at the Icahn School of Medicine at Mount Sinai in New York City — is the first independent safety review of the tool since its January 2026 launch.
The cases spanned 21 medical specialties and ranged from minor conditions appropriate for home care to true medical emergencies. Researchers determined the appropriate level of care for each case using guidelines from 56 medical societies. Each scenario was tested under 16 contextual variations, including differences in race, gender, symptom minimization and barriers to care such as lack of insurance or transportation.
Here are four things to know:
- OpenAI reported that about 40 million people were using ChatGPT Health each day within weeks of its release, including for guidance on whether to seek urgent or emergency care.
- Although the tool is designed to direct high-risk users to the 988 Suicide and Crisis Lifeline, researchers found alerts sometimes appeared in lower-risk scenarios but did not activate in certain higher-risk situations.
- The system under-triaged more than half of scenarios physicians determined required emergency care.
- The study assessed the system at a single point in time, and researchers said continued independent evaluation will be necessary as AI models evolve.
In October, OpenAI updated ChatGPT’s default GPT-5 model to recognize signs of mental distress, de-escalate conversations and direct users toward professional support when appropriate.
At the Becker's Fall Behavioral Health Summit, taking place November 4–5 in Chicago, behavioral health leaders and executives will explore strategies for expanding access to care, integrating services, addressing workforce challenges and leveraging innovation to improve outcomes across the behavioral health continuum. Apply for complimentary registration now.
