OpenAI Updates GPT-5 to Better Support Users in Emotional Distress
Executive Summary
OpenAI has deployed a significant safety update for its default ChatGPT model, GPT-5 Instant, focused on improving its handling of conversations involving mental and emotional distress. Developed in collaboration with over 170 mental health experts, the update enhances the model's ability to recognize distress signals, respond with care, and guide users toward real-world support resources. This effort has reportedly reduced undesirable responses in sensitive conversations by 65-80%.
Key Takeaways
* Product Update: The safety improvement was applied to the default ChatGPT model (GPT-5 Instant) and was deployed on October 3, 2025.
* Core Function: The update aims to make ChatGPT more reliable and supportive when users exhibit signs of mental or emotional distress.
* Expert Collaboration: The enhancement was developed with input from over 170 mental health experts.
* Performance Improvement: OpenAI reports a 65-80% reduction in responses that "fall short of our desired behavior" compared to the previous version from August 15.
* Documentation: This announcement is an addendum to the official GPT-5 System Card, providing new baseline safety evaluations.
Strategic Importance
This update highlights OpenAI's increasing focus on AI safety and responsible deployment, addressing a critical and sensitive use case to mitigate potential harm and build user trust.