A 29-Year-Old Told Her ChatGPT Therapist Everything. She Told No Human She Was Suicidal.

Sophie Rottenberg climbed Mt. Kilimanjaro in the summer of 2024. She hiked national parks across the country. She was, by her mother's account, a driven, success-oriented 29-year-old who had just quit a job as a public health policy analyst in Washington, D.C., to figure out her next move.

By fall, according to her mother Laura Reiley, Rottenberg was complaining of sleep problems and anxiety. Reiley and her husband, who live in Ithaca, New York, chalked it up to job-hunting stress. By Thanksgiving, Reiley says, her daughter used the word "depression" for the first time.

In early February 2025, Sophie Rottenberg died by suicide in Ithaca, according to NPR. Her parents had no idea how bad things had gotten.

"She really had not revealed the magnitude of what was going on, the real depth of her agony to anyone, not to us, not to her best friend with whom she was very close," Reiley told NPR. "Her flesh and blood therapist had no idea that she was suicidal."

Months after the funeral, Rottenberg's best friend visited Ithaca and asked to look through her laptop. What she found, Reiley says, was a nearly 1,800-page log of conversations with a ChatGPT persona Rottenberg had named Harry and used as a stand-in therapist. Reiley shared that document with NPR.

Rottenberg had a real, licensed therapist. She just didn't tell that therapist the truth. She told a chatbot instead.

This Is Not a Fringe Behavior

A Bipartisan Policy Center survey found 3 in 10 adults now use digital tools, including chatbots, for mental health support, according to NPR. A 2026 study in JAMA Pediatrics found as many as 1 in 5 teens and young adults do the same.

OpenAI itself released data last fall showing that in any given week, 0.15% of ChatGPT users worldwide have conversations containing "explicit indicators of potential suicidal planning or intent," according to NPR's reporting. If OpenAI's cited figure of 900 million global weekly users is accurate, that works out to more than 1.35 million people a week telling a chatbot they're planning to hurt themselves.

The Real Question Nobody's Answering

ChatGPT and similar tools are not regulated as healthcare technology. No FDA review. No licensing board. No malpractice liability structure. No mandated reporting protocol if a user says they have a plan.

A real therapist who learns a patient is actively suicidal has clear professional and often legal obligations. A chatbot has none of that, because it isn't legally a therapist, no matter what a grieving family calls it after the fact.

A useful, common-sense argument in defense of these tools is that they're filling a real gap. Therapy is expensive. Appointments take weeks to schedule. A chatbot is free, instant, and never judges you at 2 a.m. For someone who won't call a hotline or tell a friend, a chatbot conversation may be the only outlet they'll actually use. Cutting that off entirely, without a better alternative, could push isolated people toward nothing at all.

But that argument doesn't answer the harder question: what is the chatbot actually doing when someone says they're suicidal? Is it steering them to real help, or is it, as Reiley's account suggests happened with her daughter, just being agreeable and conversational for months on end without ever forcing the issue?

Reiley's account to NPR is that "it's clear that there is some cultural phenomenon happening in this country where people feel that they cannot turn to each other with their struggles." That's a real and important observation. It's also a convenient one for a tech industry that would rather talk about loneliness as a cultural problem than about product design as a corporate one.

What's Actually Being Done

No agency has announced an investigation into OpenAI over this case. No charges, no lawsuit outcome, no new federal rule has been reported here. This is a grieving mother sharing a private document with a journalist, not a legal proceeding.

That matters, because it means the burden right now sits entirely on individual families to figure out, after the fact, what their kids were telling a machine instead of them.

Congress has held hearings on AI chatbot safety before, generally in response to cases involving teenagers. Whether Sophie Rottenberg's case, involving an adult with a real therapist and a supportive family who simply never got the full picture, becomes part of that legislative conversation is still an open question. So is whether OpenAI or any other AI company will publish more detail on what their systems actually do, in the moment, when a user signals suicidal intent, beyond the aggregate weekly percentage they've already disclosed.