OpenAI taps pediatricians and psychologists for AI well-being council

OpenAI is formalizing a group of outside experts to answer a question most tech companies dodge: what does a healthy relationship with an AI actually look like? The newly announced Expert Council on Well-Being and AI brings together heavy hitters in psychology, psychiatry, and human-computer interaction to guide how ChatGPT and Sora behave, especially around vulnerable users.

You can see the fingerprints of this group already. OpenAI had been informally consulting several members earlier this year when building its parental controls and crafting the delicate language used to notify parents if a teen appears to be in distress. The council’s feedback directly shaped the tone of those messages, pushing for something that feels caring rather than corporate. The formalized roster now includes specialists like David Bickham from Boston Children’s Digital Wellness Lab, who studies social media’s impact on youth mental health, and Mathilde Cerioli, chief scientist at everyone.AI, who focuses on how AI intersects with a child’s cognitive development.

This isn’t just a press release talking shop. The council kicked off with an in-person deep dive last week, embedding members directly with the product teams they’ll be advising. Recurring meetings will tackle thorny scenarios, such as how an AI should navigate complex emotional situations and what guardrails actually support well-being rather than just paying lip service to it. The initial discussions reportedly grappled with defining well-being itself—a philosophical debate that has immediate practical stakes when you’re building a product used by millions of teens.

Alongside this council, a separate multidisciplinary subset of clinicians within the company’s Global Physician Network is stress-testing ChatGPT’s responses in real-world scenarios spanning crisis intervention and pediatrics. It’s a dual-layer approach: one group shaping the philosophical roadmap for well-being, another validating whether the model actually behaves in a clinically sound way. The real test is whether this advice translates into shipped features that respect user autonomy while offering meaningful support—and whether the company can resist the gravitational pull of engagement metrics when they conflict with mental health best practices.