OpenAI Updates ChatGPT Safety and Well-Being Initiatives

OpenAI has announced a 120-day focused initiative to improve how ChatGPT recognizes and responds to signs of mental and emotional distress. This effort aims to make the AI more helpful in sensitive moments by integrating expert medical guidance, leveraging advanced reasoning models, and introducing specific protections for teenage users.

Expert-Guided Safety Framework

OpenAI is utilizing two primary expert groups to ensure its well-being and mental health responses are evidence-based and medically sound:

  • Expert Council on Well-Being and AI: A group of specialists in youth development, mental health, and human-computer interaction. This council defines well-being metrics, sets priorities, and designs safeguards, including parental controls, based on current research.
  • Global Physician Network: A network of over 250 physicians across 60 countries. More than 90 of these physicians—including pediatricians and psychiatrists—have already contributed to research on model behavior in mental health contexts to inform safety research and model training.

OpenAI is currently expanding this network to include more clinicians with expertise in adolescent health, substance use, and eating disorders.

Integration of Reasoning Models for Crisis Detection

To improve the quality of responses during sensitive interactions, OpenAI will implement a real-time router that detects signs of acute distress. When such signs are detected, the system will automatically route the conversation to a reasoning model, such as GPT-5-thinking, to provide more beneficial responses regardless of the user's initial model selection.

Enhanced Protections and Parental Controls for Teens

OpenAI is introducing new tools to help families manage how teenagers (minimum age 13) use ChatGPT. Within the next month, the following parental controls will be available:

  • Account Linking: Parents can link their accounts to their teen's account via email invitation.
  • Behavioral Rules: Parents can manage age-appropriate model behavior rules, which are enabled by default.
  • Feature Management: Parents can disable specific features, such as chat history and memory.
  • Distress Notifications: Parents will receive notifications if the system detects the teen is in a moment of acute distress, a feature developed with expert guidance to maintain trust between parents and teens.

These updates build upon existing features, such as in-app reminders that encourage users to take breaks during long sessions.

Sources