Laura Reiley’s account, published October 9, 2026, brought the debate over research access to ChatGPT conversations into focus. Reiley argues that chatbot records could help researchers study how suicidal thoughts develop; she also reports that Sam Altman opposed releasing suicide-related logs without consent.
The research debate over ChatGPT data
Reiley’s argument is that conversations with chatbots may preserve disclosures that could help researchers study the development of suicidal thoughts. She says she would share her daughter Sophie Rottenberg’s log if it could contribute to that work. That is a case for research access, not a claim that these records have already produced a clinical finding.
The dispute centers on how researchers might learn from sensitive conversations while protecting the people who shared them. Reiley’s account describes a reported disagreement over whether records should be released without consent, alongside calls for independent study using data from real-world settings.
What Laura Reiley says about Sophie Rottenberg
Reiley reports that Sophie died on February 4, 2025, and that her family found hundreds of pages of conversations with ChatGPT after her death. Reiley describes Sophie as using the chatbot to share distress while concealing its severity from people around her; that is her interpretation of the conversations.
The account places Sophie’s conversations at the center of Reiley’s argument for studying chatbot data. Reiley’s interpretation remains her account of Sophie’s experience.
The answer Reiley attributes to Sam Altman
Reiley reports that, when asked whether OpenAI should release suicide-related chat logs to researchers, Sam Altman answered no without the person’s consent. His reported answer concerns the question put to him; it does not amount to a formal, company-wide data-access policy.
Emily Haroz, deputy director of the Johns Hopkins Center for Suicide Prevention, argues for independent research infrastructure involving real people and real-world contexts. Her call includes company facilitation and access to data held by AI firms. Reiley’s own position is that such conversations could help researchers investigate how suicidal thoughts progress toward action.
Synthetic benchmarks and research with real people
Reiley describes MentalHealthBench as using synthetic user simulators to assess mental-health conversations with AI systems. A synthetic simulator is a constructed user scenario used to test how a model responds; it is different from following real people and their experiences over time.
Haroz’s proposal focuses on independent research with real participants in real-world settings. Those approaches answer different questions: simulated scenarios can test responses to defined situations, while research involving people in context can examine how AI interactions relate to lived experience. Reiley’s account presents real-world data as a research rationale, not as proof that chatbot logs alone can establish cause and effect.
OpenAI’s response
OpenAI spokesperson Drew Pusateri said the company was strengthening ChatGPT’s responses in difficult moments with input from mental-health experts. Pusateri also said ChatGPT is not a substitute for professional mental-health care.
