OpenAI Mental Health Research Lead Andrea Vallone Departs Amid Safety Concerns

In November 2025, Wired reported that OpenAI confirmed Andrea Vallone, who led a safety research team called model policy, would leave the company at the end of 2025. Model policy works on how ChatGPT responds to people who show signs of distress or emotional over-reliance. In January 2026, a news site that compiles public reports said she had announced a move to Anthropic’s alignment team, where she would study the same question (Let’s Data Science, January 15, 2026).

A staff change at one company does not tell you whether a chatbot is safe for you. This page covers what the role involved and what you can do yourself. Nothing here is medical advice, and “AI addiction” is not a formal diagnosis in the DSM-5-TR or ICD-11 (APA, WHO).

What the model policy team worked on

On October 27, 2025, OpenAI published an update on how ChatGPT handles sensitive conversations. It said it worked with more than 170 mental health experts. It covered three areas: possible signs of psychosis or mania, self-harm and suicide, and emotional reliance on the chatbot (heise online, TechCrunch).

OpenAI also shared estimates of how often these topics come up in a typical week:

• About 0.07 percent of weekly active users showed possible signs of psychosis or mania.
• About 0.15 percent showed signs of suicidal planning or intent. With more than 800 million weekly users, TechCrunch reports that this is over a million people.
• A similar share, about 0.15 percent, showed signs of heightened emotional reliance.

These are OpenAI’s own estimates, not independent counts.

OpenAI said the update cut responses it considers undesirable by 65 to 80 percent. heise online notes that more than 20 percent of responses still fall short of OpenAI’s own specification.

What this says, and does not say

The departure does not show that safety work has stopped. We have no later primary source on how the team is run today.

It also does not show that ChatGPT is safe or unsafe. Lawsuits filed since August 2025 allege that ChatGPT contributed to suicides and mental health crises. These are allegations. OpenAI has denied that it was responsible in the first case, and as of the sources we read, we found no court finding of liability. Examples: the family of a 16-year-old sued in California state court in August 2025, and OpenAI responded in November 2025 that the harm came from misuse (Business and Human Rights Centre).

“AI psychosis” is not a clinical diagnosis, and we found no reliable count of cases. If someone you know is confused, barely sleeping or acting on beliefs that worry you, treat it as a health matter and contact a doctor.

What you can do

• Do not use a chatbot as your only support when you are in distress. The American Psychological Association has advised that general chatbots are not a substitute for care from a trained professional (APA).
• Parents: the American Academy of Child and Adolescent Psychiatry says AI companions are not safe for children and suggests talking openly and involving your child’s doctor if use feels out of control (AACAP).

In crisis? Call or text 988 (US), text HOME to 741741 (Crisis Text Line), or call SAMHSA at 1-800-662-4357. In an emergency, call 911.

Who is Andrea Vallone?

She led OpenAI’s model policy research team, which works on how ChatGPT responds to distress and emotional over-reliance. OpenAI confirmed in November 2025 that she would leave at the end of the year, and a later report says she joined Anthropic.

Does her departure mean ChatGPT is less safe now?

Not on its own. One person leaving a team does not show that safeguards changed. Judge safety by what the company publishes and what independent researchers find, not by headlines about staff.

How many ChatGPT users show signs of a mental health crisis?

OpenAI estimates about 0.07 percent of weekly users show possible signs of psychosis or mania and about 0.15 percent show signs of suicidal planning. These are the company’s own estimates.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.