OpenAI’s ChatGPT now includes an optional Trusted Contact feature intended to connect people experiencing serious suicidal thoughts or self-harm concerns with someone they trust. But the feature must be enabled in advance, and users may not be told about it when they first create an account. The tool, introduced in May, allows users to designate a contact who may be alerted if ChatGPT detects signs of a serious safety risk. It is available through the service’s settings and is not activated automatically. In testing, a new user who created a ChatGPT account on a phone received no prompt about Trusted Contact. When asked directly about the feature, ChatGPT initially said it was not aware of a general option for people with mental health concerns. After being challenged, it acknowledged that the feature exists and explained that it applies to serious safety concerns rather than ordinary stress. The system also clarified that nothing happens if a user has not enrolled beforehand. Without a designated contact, ChatGPT cannot notify another person on the user’s behalf during a crisis. OpenAI spokesperson Gaby Raila said the feature is not yet available during voice conversations, though the company is working on adding it. In text chats, ChatGPT may suggest setting up a Trusted Contact if it detects suicidal thoughts and significant emotional distress, Raila said. The feature’s limitations matter because ChatGPT has become a place where many people discuss mental health. Estimates based on data released by OpenAI indicate that more than 1 million people each week discuss suicide or self-harm with the chatbot. Suicide prevention specialists say those conversations underscore the importance of directing users toward real human support rather than relying on an automated system. Holly Wilcox, director of the Center for Suicide Prevention at the Johns Hopkins Bloomberg School of Public Health, said the feature would be most useful if offered to users before they reach an emergency. Psychologist Ursula Whiteside, chief executive of the suicide-prevention nonprofit Now Matters Now, tested the enrollment process and found the steps generally straightforward. But she said a person who is extremely upset could have more difficulty following them. Whiteside said people in peer-support groups have expressed interest in choosing a trusted person while thinking clearly, so that support is available when suicidal thoughts become overwhelming. Human connection and a sense of belonging are widely viewed by suicide-prevention professionals as important protective factors. Jennifer Lockman, who runs a suicide-prevention clinic at the University of Alabama, said some people considering suicide may resist contacting others because they believe they are a burden. She questioned whether a chatbot can overcome those barriers or replace the judgment of a trained mental health professional helping someone identify appropriate support. Still, Lockman described the tool as another possible path to human connection. She said more research is needed before anyone can determine whether it prevents deaths. OpenAI says Trusted Contact is one part of a broader set of safeguards, including encouraging users to take breaks and directing people to crisis resources. The company says it is continuing to refine when and how the feature is offered, in consultation with psychologists and researchers. People in the United States who are experiencing a mental health crisis or thinking about suicide can call or text 988 to reach the Suicide and Crisis Lifeline. In an immediate emergency, call 911 or go to the nearest emergency department.