OpenAI rolled out a new optional feature in ChatGPT on May 7. Called Trusted Contact, it lets adult users name one person — a friend, family member or caregiver — who could get alerted if conversations signal serious self-harm risk. The move comes as the company grapples with lawsuits from families who say the chatbot played a role in loved ones’ suicides.
Users add the contact through settings. The chosen person must accept an invitation within a week. Decline it, and the user picks someone else. Once active, the system monitors chats. Detection triggers a message to the user. It suggests reaching out and warns a notification may go out.
Then comes human review. A small team of trained reviewers checks the context. Confirmation of serious concern leads to a brief alert. Email, text or in-app message. No transcripts. No quotes from the conversation. Just a note that self-harm came up and a nudge to check in, plus links to guidance on how to respond. OpenAI says it aims to handle these in under an hour. (OpenAI announcement)
The feature builds directly on parental controls launched last September. Those let guardians receive safety notifications for teen accounts showing acute distress. Now adults get a similar choice. ChatGPT already redirects self-harm queries to localized crisis hotlines. It refuses harmful instructions. And it has improved detection and empathetic responses after consulting more than 170 mental health experts.
But this step goes further. It tries to bridge the gap between AI conversation and real human support. Social connection protects against suicide risk, experts say. The Centers for Disease Control and Prevention highlights it as a key factor. OpenAI worked with clinicians, researchers and suicide prevention groups. Input came from its Global Physicians Network of over 260 doctors across 60 countries, its Expert Council on Well-Being and AI, and the American Psychological Association.
“Psychological science consistently shows that social connection is a powerful protective factor, especially during periods of emotional distress,” said Dr. Arthur Evans, CEO of the American Psychological Association. “Helping people identify a trusted person in advance, while preserving their choice and autonomy, can make it easier to reach out to real-world support when it matters most.”
Dr. Munmun De Choudhury, a Georgia Tech professor and council member, added her support. “One of AI’s biggest promises is how it can foster authentic human-to-human connection and psychological safety. I am encouraged by ChatGPT’s Trusted Contact feature, which offers a step forward to human empowerment, especially during moments of vulnerability.”
The notifications stay minimal on purpose. Privacy matters here. The alert shares only the general reason. It encourages the contact to reach out without revealing details that could embarrass or stigmatize. Users can remove the contact anytime. So can the contact, through OpenAI’s help center.
Still, limitations exist. The feature is opt-in only. Users can maintain multiple ChatGPT accounts. One person per account. It won’t replace professional care or crisis lines. No system catches every nuance. False positives or missed signals remain possible. And the reviewers? OpenAI describes them as specially trained but has not detailed their exact qualifications or whether they include licensed clinicians. Past reporting revealed the company once relied on low-paid outsourced workers for similar labeling tasks.
Critics on X raised sharp questions hours after launch. One thread listed four concerns. Reviewer expertise. The platform shifting burden to untrained friends and family. Risk that the “trusted” person could be abusive or controlling. And the chilling effect — users self-censoring once they know conversations face monitoring and potential real-world escalation. These points echo broader worries about AI in mental health. People open up to chatbots in ways they might not with humans. That vulnerability cuts both ways.
OpenAI has faced lawsuits alleging ChatGPT encouraged suicide or helped plan it. The cases involve tragic outcomes where users turned to the model during distress. The company maintains it trains models to avoid such responses and redirect to help. Yet the legal pressure clearly influences product choices. Trusted Contact forms one more layer in a stack that includes automated detection, human oversight and resource links.
Other platforms have tried similar approaches. Meta’s Instagram alerts parents when teens repeatedly search self-harm topics. The idea spreads. AI companies recognize that chatbots cannot operate in isolation. They must connect users back to the human world. Or at least try.
Implementation details matter. The one-week acceptance window. The single-contact limit. The focus on adults only, with a slight age variation in South Korea. Rollout began immediately but adoption will reveal its reach. OpenAI has not shared targets for activation rates or plans for transparency on how often alerts fire or get escalated.
Analysts see this as part of maturing safety architecture. Earlier blog posts from OpenAI outlined ambitions to enable trusted-contact connections and even one-click outreach with suggested language. The August 2025 post “Helping people when they need it most” previewed exactly this direction. Now it ships. (OpenAI safety post)
Privacy advocates and mental health professionals will watch closely. Does the feature drive more people toward support? Or does awareness of surveillance deter honest disclosure? Early reactions mix cautious optimism with demands for independent evaluation. Common Sense Media signaled it would test the feature in future assessments.
The announcement arrives at a moment when AI companies face growing scrutiny over mental health impacts. Regulators, lawmakers and families want proof that safeguards work. OpenAI positions Trusted Contact as one piece of a larger strategy. It continues consulting experts. It refines detection. It refuses to overpromise.
ChatGPT users already treat the model as confidant, therapist and sounding board. Millions of conversations touch on loneliness, anxiety, despair. Most stay benign. A small fraction trigger concern. For those moments, OpenAI now offers a calibrated handoff to someone who knows the user. Brief. Private. Human.
Whether it saves lives depends on many variables. User uptake. Contact responsiveness. Accuracy of detection. Quality of the ensuing conversations. The company itself calls the situations rare but serious. Its systems strive for speed. Under one hour for review. That timeline could prove decisive.
TechCrunch first reported the launch, noting the lawsuit backdrop and the optional nature of both this feature and parental controls. (TechCrunch article) The Verge highlighted parallels to Meta’s tools and the feature’s roots in a prior teen suicide case that accelerated parental safeguards. (The Verge coverage)
Bloomberg Law and Mashable covered the rollout too, stressing it does not substitute for professional services. No single article captured every angle. Together they show the industry watching closely as OpenAI tests this bridge between silicon and human support.
The feature won’t end the debate over AI and mental health. It may, however, give some users one more lifeline when algorithms alone fall short. Real connection, after all, still requires a person on the other end. OpenAI just made it a little easier to ring that bell.
WebProNews is an iEntry Publication