Millions of teenagers are already using ChatGPT as a first-response mental health resource, with zero clinical oversight built into the product.

AI companies have been guessing at emotional safety — and teens are paying the price

Until now, the guidelines shaping how AI responds to users in emotional distress were written by engineers, not clinicians. That gap between product logic and psychological best practice has left the most vulnerable users exposed to responses that are either dismissive or dangerously over-affirming.

A council of psychologists now sits inside OpenAI’s product decisions

Expert Council on Well is a standing advisory body of leading psychologists, clinicians, and academic researchers assembled by OpenAI to directly influence how ChatGPT is built and trained. Members review real behavioral patterns, flag risks in model responses, and set research priorities that feed back into product updates. The output is not a report that sits on a shelf — it is embedded guidance that shapes what ChatGPT actually says when a user is struggling.

The professionals watching this most closely work in schools, clinics, and policy offices

• School counselors managing student caseloads who need AI tools that will not contradict evidence-based intervention protocols when students go home and open ChatGPT
• Clinical researchers studying AI’s effect on adolescent mental health who now have a formal channel to influence a product used by hundreds of millions of people
• EdTech policy leads at district and state level who must decide whether AI tools are safe to recommend before the next budget cycle

This is the first time a major AI lab has made clinical expertise structurally part of the product team rather than a one-time audit.

Regulators were already moving — OpenAI moved first

The EU AI Act and multiple US state-level proposals specifically target AI systems that interact with minors around sensitive topics, with compliance deadlines hitting in 2025 and 2026. Expert Council on Well-Being and AI positions OpenAI ahead of that wave, but it also raises the bar every competitor now has to clear.

What this council is built to do

• Review ChatGPT’s emotional response patterns against clinical standards
• Identify model behaviors that could worsen mental health outcomes for teens
• Set research priorities that feed directly into training updates
• Publish findings that inform the broader AI safety field

Pricing not listed — check our directory.

An honest limit worth naming

Advisory councils can be ignored — there is no public mechanism yet to verify that the council’s recommendations are actually implemented in model updates.

Other efforts in this space

Google DeepMind has published safety research on AI and adolescent well-being, but has not created a standing clinical body with direct product influence. Anthropic’s Constitutional AI approach bakes in ethical constraints at training time, which addresses different risks than real-time clinical guidance does.

Clinical oversight of AI is becoming a competitive requirement, not a differentiator

This is the kind of structural shift that rewrites procurement checklists for every school district and healthcare system considering AI tools. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.