OpenAI adds Trusted Contact safeguard

ChatGPT is getting a new kind of safety backstop. OpenAI started rolling out an optional feature called Trusted Contact on May 7 that lets adult users name a friend, family member, or caregiver who can be alerted if a conversation appears to involve serious self-harm risk. The point is simple — if someone is in real trouble, the chatbot should have a way to pull an actual person into the loop. That is a meaningful shift, because most AI safety features stop at warnings, crisis resources, or refusal messages. (openai.com) ### What is Trusted Contact? It is an opt-in setting inside ChatGPT for adults. A user chooses one person they trust, enters that person’s contact details, and that person has to explicitly accept the invitation before anything is active. If the invite is declined or ignored, the user has to pick someone else. This is not on by default, and it is not a general “share my chats” feature. (openai.com)ually send an alert? Not when someone is merely sad, venting, or asking broad questions about mental health. OpenAI says the trigger is much narrower — cases where automated systems and trained reviewers think the enrolled user may have discussed harming themselves in a way that signals a serious safety concern, including possible suicide risk. In those cases, ChatGPT may notify the trust(openai.com)ne here; a human review layer sits between the flag and the alert. (openai.com) ### Why add a human contact at all? Because a chatbot can de-escalate, but it cannot physically show up, call you, or notice that something feels off in the room. OpenAI has already been surfacing crisis resources and nudging distressed users toward real-world help. Trusted Contact adds a more direct bridge to offline support — basically, one step beyond “here is a hotline” and closer to “someone you know should check on you now.” (openai.com) ### Is this replacing emergency services? No. OpenAI is framing this as a support feature, not emergency intervention. That matters because the company has also said it does not currently refer self-harm cases to law enforcement, partly because ChatGPT conversations are unusually private. Trusted Contact is a different model — keep the response personal and voluntary, rather than turning a mental-health crisis into a police matter by default. (openai.com) ### Why now? This has been building for a while. OpenAI published a broader mental-health update in March that previewed trusted contacts, parental controls, and better distress detection. Parents of teen users can already receive limited safety notifications in some cases. Trusted Contact extends that idea to adults, but as a voluntary feature rather than an account-level control imposed by a guardian. (openai. ([openai.com) is the hard part here? False positives and trust. If alerts fire too easily, users will feel watched. If they fire too rarely, the feature misses the people it is meant to help. That is why OpenAI keeps emphasizing the two-step system — automated detection plus trained reviewers — and why enrollment requires consent from both sides. The design is trying to thread a needle between intervention and privacy. (openai.com) ### Does this change what ChatGPT is becoming? Yes — a little, but in an important way. ChatGPT is no longer just being treated as a tool for answers and brainstorming. OpenAI is building for the reality that people use it during lonely, emotional, and sometimes dangerous moments. Trusted Contact does not make the chatbot a therapist. But it does make the product more willing to hand off to an actual human when the stakes get high. (openai.com) ### Bottom line Basically, OpenAI just added an emergency-adjacent feature for adults who want one: if ChatGPT sees signs of possible suicide risk, it can now try to reach a real person you chose in advance. That is a bigger deal than it sounds, because it turns AI safety from a pop-up message into a human escalation path. (openai.com)