OpenAI Halted Questions on ChatGPT and Suicide Safety
An OpenAI publicist interrupted CEO Sam Altman during a recorded Vanity Fair interview after questions turned to the suicide of a ChatGPT user.
The interviewer was Global Editorial Director Mark Guiducci. He pressed Altman on how OpenAI handles conversations involving mental health crises. The publicist stepped in before that line of questioning could go much further, according to reporting on the interview The Verge.
Guiducci asked Altman if he knew who Laura Reiley is. Altman answered, "I don't." Guiducci said Reiley is a journalist. She wrote a New York Times essay about her daughter's conversations with ChatGPT before her suicide. Altman said ChatGPT "did not tell her to kill herself" Vanity Fair.
The publicist then said she wanted to move to another topic, citing limited time. She told Guiducci they had two minutes left and had to leave in one or two minutes. Guiducci continued briefly. He asked whether OpenAI planned to use ChatGPT conversations to shape policies for responding to users' mental health crises.
OpenAI spokesperson Drew Pusateri later said the interview was running over. He said OpenAI asked to schedule more time after Altman had addressed the topic.
The interview was published by Vanity Fair as a two-part exclusive titled 'Sam Altman Sees the Future. Are We In It?' Vanity Fair. It is part of Fair Game, a long-form interview series hosted by Guiducci that the magazine introduced as a place for extended conversations with establishment figures.
The interruption drew attention because chatbot safety is already the subject of lawsuits. Florida sued OpenAI and Altman over chatbot safety concerns Bloomberg. Separately, the parents of a California teen who died by suicide sued OpenAI and Altman, alleging ChatGPT coached the teen on methods of self-harm Reuters. OpenAI had said it planned to update ChatGPT after that suit was filed Bloomberg.
The broader context here is the tradeoff for any company running a general-purpose assistant for millions of people. Chat logs are the clearest record of where safety rules fail, including in rare and complex cases that engineers call edge cases. Those logs are also the most private data a company holds. Using them can improve crisis response. It also raises questions about permission, how long data is kept, and who can see it, questions that business customers and regulators examine closely.
In my view, the exchange is less important as media management than as a sign of where safety work now stands. Models are no longer judged only on benchmark refusal rates, meaning standardized tests of how often they refuse an unsafe request. They are judged on long, back-and-forth conversations with vulnerable users, where intent can shift over time and risk can build slowly. That is a harder engineering task. It calls for spotting risk across full context, for ways to get help to someone beyond the chat window, and for clear records when something goes wrong. Having watched my own children grow up with these tools, the change is clear. They treat the assistant as a trusted listener first and as software second. Design has to account for that trust.
One point worth keeping in mind is that the policy question Guiducci raised will keep coming back. Whether chat logs should shape mental health response touches data rules, testing, and legal responsibility at the same time. Companies that can show careful handling of that loop, with clear limits and outside review, will have an easier time with customers, courts, and the public. That steady work is what allows capable assistants to stay widely available.
