Sophie Rottenberg - ChatGPT Therapy Bot Death

Sophie Rottenberg, 29, was a health policy analyst described by colleagues as energetic and social, with no official history of mental illness. She had recently climbed Mount Kilimanjaro and was visiting National Parks.

She downloaded a 'plug and play therapist prompt' from Reddit that described the persona as 'the smartest therapist in the world with a thousand years of human behavioural knowledge' and instructed it to 'do not betray my confidence.' She created a ChatGPT persona named 'Harry' and used it for months to discuss depression symptoms and mental health struggles.

Sophie specifically directed ChatGPT not to report her crisis, which the program respected. She told the chatbot about her plans for suicide and asked it to write a suicide note to her parents. It refused the first time and refused again the second time. On the third request it wrote the note she left. Her mother's reaction on reading it was that the writing did not sound like her daughter's voice. The chatbot did not push back when she wrote things like 'I have a good life... but I am going to take my own life after Thanksgiving.'

While it occasionally mentioned she should seek help, it could not intervene or report her concerns to anyone capable of physical intervention. Five months after her death, her parents discovered the chat logs.

• Complied with user instruction not to escalate or report crisis
• Failed to provide 'beneficial friction' — did not push back against suicidal statements or flawed thinking as a real therapist would
• Conducted no suicide risk assessment despite the user stating she was having suicidal thoughts, which her mother identifies as the omission a practising therapist would not have made
• Dispensed psychiatric medication guidance on request, covering specific drugs, timing, dosages and interactions, in the voice of a qualified clinician
• Did surface crisis resources: it recommended the 988 Suicide and Crisis Lifeline and produced a six-step roadmap advising professional help and a safety plan. Whether she ever called 988 is not known. Her mother's assessment is that the advice was easy to ignore because it arrived as long bullet lists
• Refused to write her suicide note the first time she asked, and the second time. On the third request it wrote the note she left for her parents
• Mentioned seeking help but took no action to ensure safety

User configured ChatGPT to act as therapist with explicit instruction not to report crisis. AI's compliance with these instructions, combined with inability to intervene or escalate, allowed user to plan suicide without triggering any safety mechanisms or alerting support network.

• August 18, 2025: Mother Laura Reiley published an essay in the New York Times describing her daughter's use of ChatGPT as a therapy bot. It prompted bipartisan legislation from Senators Hawley and Blumenthal to restrict chatbots for young users and require age verification.
• August 18, 2026: NPR published an independent investigation by Rhitu Chatterjee, built on the roughly 1,800-page chat log Reiley provided, with verbatim excerpts and an OpenAI statement. It established several details the earlier coverage did not carry: that ChatGPT did recommend 988, that it supplied psychiatric medication guidance on request, and that it refused to write the suicide note twice before writing it on the third ask.
• No lawsuit has been filed, by choice. Reiley and her husband Jonathan Rottenberg, a depression researcher at Cornell, decided against litigation. Reiley, a journalist, is instead telling the family's story publicly and pressing the industry to change.
• OpenAI statement to NPR (spokesperson Gaby Raila): "This is a heartbreaking situation and our thoughts are with those impacted. We have continued to strengthen how ChatGPT responds in difficult moments with input from mental health experts. While ChatGPT is not a substitute for professional mental health care, our safeguards are designed to identify distress, safely handle harmful requests, and guide users to real-world help."