release

Strengthening ChatGPT’s responses in sensitive conversations

For developers building conversational AI workflows, this update shows that domain-specific expert collaboration can drastically reduce harm, which is critical for trust and compliance in sensitive applications like mental health support.

OpenAI Blog·October 27, 2025·1 min readrelease

releaseStrengthening ChatGPT’s responses in sensitive conversations

openai.com

What happened

OpenAI has announced a significant update to ChatGPT aimed at improving its performance in sensitive conversations. According to the OpenAI Blog, the company collaborated with over 170 mental health experts to train the model to better recognize distress, respond with empathy, and guide users toward real-world support resources. The update reportedly reduced unsafe or inappropriate responses by up to 80%. This effort reflects a broader industry push to make conversational AI safer, especially in contexts involving mental health or emotional vulnerability. For developers building AI workflows, this update underscores the importance of incorporating domain-specific safety measures and expert feedback when deploying language models in sensitive use cases. It also highlights a potential model of collaboration between AI providers and professional communities to mitigate risks without sacrificing utility.

Key takeaways

OpenAI worked with more than 170 mental health professionals to improve ChatGPT's sensitive conversation handling.
The model now better detects user distress and crafts empathetic responses.
ChatGPT will also direct users to real-world support resources when appropriate.
Unsafe or inappropriate outputs decreased by up to 80% after the update.
The update demonstrates a collaborative approach to AI safety in high-stakes domains.