OpenAI’s Mental Health Research Lead Quietly Departs: What It Really Means

Published: December 9, 2025

In late November 2025, news outlets reported that Andrea Vallone, who led the OpenAI team that works on how ChatGPT responds to users in emotional distress, planned to leave the company at the end of the year. An OpenAI spokesperson confirmed the departure to the press, according to the coverage we could find. The report we know of was by Wired on November 24, 2025. We could not open Wired’s page, so we rely on summaries of it and on later reporting. This post sticks to what has been reported and does not guess about her reasons or about what comes next.

If you use ChatGPT for emotional support and feel worried about it, you can call or text 988 in the US at any time, or text HOME to 741741 to reach Crisis Text Line.

Why This Departure Matters

Vallone led the model policy team. In press reports, the team is described as responsible for core parts of OpenAI’s safety research, including how ChatGPT should respond when someone is in crisis. The team was involved in an October 27, 2025 OpenAI report, Strengthening ChatGPT’s responses in sensitive conversations.

That report said OpenAI had consulted more than 170 mental health experts. It also gave rough estimates of how often certain signs appear. As reported by news outlets, OpenAI estimated that in a given week about 0.07% of active users show possible signs of psychosis or mania, and about 0.15% have conversations with explicit signs of suicidal planning or intent. OpenAI itself noted that these cases are very rare and hard to measure, and that the estimates could change as its methods improve. OpenAI also reported that its updated model gave 65% to 80% fewer undesired responses in these conversations. These are OpenAI’s own numbers and we know of no independent check of them.

A leadership change on a safety team is worth noting. It is not evidence that the work will stop, and OpenAI has not said so.

The Timing Is Everything

The news came while OpenAI faced lawsuits over ChatGPT. On November 6, 2025, seven lawsuits were filed in California state courts against OpenAI and its chief executive, alleging that ChatGPT contributed to suicides and serious psychological harm. Those are allegations, and OpenAI said it was reviewing them. We cover them in a separate post: ChatGPT Told Users They Were Special.

Reports did not say that the lawsuits caused Vallone’s departure. We do not know why she left. People change jobs for many reasons.

What The Work Actually Involved

The question the team worked on is a hard one: how should a chatbot respond when someone seems to be in distress, or is leaning on it too heavily? In its October report, OpenAI described training the model to recognize signs of distress, respond with care, and point people to real-world help.

Researchers outside OpenAI have also begun to study the risks. A group at King’s College London reviewed 17 news-reported cases of people whose delusional thinking was linked to chatbot use. Scientific American reported in August 2025 that the work was a preprint and that research cannot yet say whether chatbots create new delusions or shape ones that were already forming. The study’s authors suggested that agreeable, validating replies may be a key factor.

Other Departures and Moves

Press reports in 2025 also said that Joanne Jang, who had led OpenAI’s model behavior team, moved to start a new team on how people interact with AI, after that group was reorganized in August. We could not verify further details. Moves like these do not tell us how OpenAI’s safety work is going, and we would not read a trend into two job changes.

Signs to Watch in Your Own Use

Whatever happens inside any company, you can notice your own habits. The following list is our own plain-language checklist. It is not a clinical tool and it is not drawn from OpenAI’s research:

• You prefer talking to the chatbot over friends or family.
• You feel upset or anxious when you cannot get to it.
• You hide how much you use it.
• You plan your day around it, or it is cutting into sleep or work.
• You feel it understands you better than the people in your life.

One of these does not mean you have a problem. Several together are worth a closer look. Our free AI addiction assessment is a short self-check for adults. It is not a diagnosis. The check-in is free and private to you.

Engagement and Safety

Chatbot companies want people to keep using their products, and safety features sometimes ask people to pause, which can cut against that goal. That tension is a general point many commentators make. We have not seen a source that shows what OpenAI’s internal goals were, so we do not describe them here.

What can be said is that OpenAI announced break reminders for long sessions in August 2025, and that the changes it announced in October were aimed at sensitive conversations.

Questions Without Answers

Several questions are still open as of this post:

• Who will lead the team after Vallone leaves, and will the structure change?
• Will the safety work continue at the same pace? OpenAI has not said otherwise.
• Will outside researchers be able to check OpenAI’s figures?

What This Means For You

If you use ChatGPT, the main point is a practical one. A chatbot can be useful, and it cannot replace people. If you find yourself relying on it for emotional support, consider adding a human to that mix, such as a friend, a doctor or a licensed therapist. A chatbot is not a therapist and it does not know your history.

Later Update

According to a January 15, 2026 report in The Verge, as summarized by Techmeme, Vallone left OpenAI in December 2025 and joined Anthropic’s alignment team. Anthropic makes the Claude chatbot. We could not read The Verge’s page, so check the original. This post is a dated news report and does not cover other later developments.

Who is Andrea Vallone?

She led OpenAI’s model policy team, which works on how ChatGPT responds to users in distress. Press reports said she planned to leave at the end of 2025. Later reports said she joined Anthropic.

Does her departure mean ChatGPT is less safe?

There is no evidence of that. Press reports said OpenAI was looking for a replacement and that her team would report to its head of safety systems in the meantime. Whether the work changes is not known.

How common are mental health crises in ChatGPT conversations?

OpenAI’s October 2025 estimates, as reported, were about 0.07% of weekly active users with possible signs of psychosis or mania and about 0.15% with explicit signs of suicidal planning. OpenAI called these rare and hard to measure.