OpenAI adds trusted contact

ChatGPT is getting a new safety layer that reaches outside the app. OpenAI started rolling out a feature on May 7, 2026 that lets adult users add one “Trusted Contact” — a friend, family member, or caregiver who can be notified if ChatGPT detects a serious self-harm risk. That matters because the weak point in a lot of AI safety work is obvious: the model can say supportive things, but it cannot physically get help to a real person. This is OpenAI trying to close that gap. ### What is the feature, exactly? It’s an optional setting for personal ChatGPT accounts. You pick one adult contact, invite them, and the feature only turns on if that person accepts. OpenAI says it’s available for adults 18+ in most places, 19+ in South Korea, and it is not enabled for Business, Enterprise, or Edu workspaces. ### How does an alert actually happen? The trigger is not just a model guessing wrong in real time. (openai.com) OpenAI says automated systems first flag a conversation that may involve self-harm or suicide in a way that signals a serious safety concern. ChatGPT then tells the user it may notify the trusted contact and encourages the user to reach out directly. After that, a specially trained human review team looks at the situation before a notification goes out. ### What does the contact receive? OpenAI’s help materials frame it pretty narrowly. The trusted contact gets a notification encouraging them to check in, not a standing feed of someone’s chats. Invitations can be sent by email, SMS, WhatsApp, or in-app messaging, but an email address is required. The company also recommends talking to the person before adding them — which sounds basic, but matters a lot when the system might pull them into a crisis moment. (openai.com) ### Why build this now? Because OpenAI has been moving in this direction for months. In October 2025, it said work with more than 170 mental health experts helped cut unsafe responses in sensitive conversations by 65% to 80%, while adding stronger distress detection and safer handoffs. Then in February 2026, it said a trusted-contact feature was coming as part of broader mental-health safety work. So this is not a random add-on — it’s the next product step in a longer safety push. (help.openai.com) ### Why not just show a hotline? Because crisis tools fail when the person in trouble does nothing. OpenAI is betting that social connection is the missing piece — basically, a real human who already knows you may be more reachable than a generic emergency prompt. The company explicitly says the feature is meant to sit alongside localized helplines, not replace them. ### What’s the catch? (openai.com) The hard part is judgment. A system like this has two bad failure modes. Miss a real crisis, and the feature looks useless. Flag the wrong conversation, and you create a privacy breach at exactly the moment a user feels most exposed. OpenAI is trying to lower that risk with opt-in enrollment, one-contact limits, acceptance requirements, and human review — but none of that makes the tradeoff disappear. (openai.com) ### Why does this feel bigger than one feature? Because it changes what “AI safety” means in product terms. Until recently, most chatbot safety work stayed inside the conversation — refuse harmful requests, suggest resources, nudge better behavior. Trusted Contact crosses into offline life. Once a chatbot can alert another person, the company is no longer just shaping words on a screen. It is making a judgment about when to involve the outside world. That is a much bigger responsibility. (openai.com) ### Bottom line This is a small setting with big implications. OpenAI is turning ChatGPT from a system that mainly responds to distress into one that may escalate distress to someone real. That could save lives. But it also means the hardest part of the job is no longer writing a careful reply — it’s deciding when a private conversation stops being private. (openai.com)