OpenAI is rolling out a new ChatGPT safety feature called Trusted Contact. It lets adult users nominate someone they trust who may be alerted if ChatGPT’s safety systems and human reviewers decide a conversation suggests a serious self-harm concern.
For UK users, this is worth understanding calmly rather than treating it as either a miracle safeguard or a privacy disaster. It is an optional feature, it is aimed at rare high-risk moments, and it sits in a sensitive space: private chats, mental health, family support and platform responsibility.
The most practical question is not “should everyone switch this on immediately?” It is “what would I want to happen if an AI system became worried about me, and who would I trust to receive that nudge?”
What Trusted Contact does
OpenAI says Trusted Contact is for adults using ChatGPT. A user can nominate one adult contact, such as a friend, family member or caregiver. That person has to accept an invitation before the feature becomes active, so it is not simply a hidden alert list.
If ChatGPT’s automated systems detect that a user may be discussing self-harm in a way that points to a serious safety concern, the system can warn the user that their Trusted Contact may be notified. It also encourages the user to reach out directly, with suggested ways to start the conversation.
OpenAI says a small team of trained reviewers then looks at the situation. If they decide the conversation indicates a serious concern, the Trusted Contact may receive a brief alert by email, text message or an in-app notification if they also use ChatGPT.
The company says the alert is deliberately limited. It is meant to explain that self-harm came up in a concerning way and encourage the contact to check in. OpenAI says it does not include the chat transcript or detailed conversation content.
Why this matters
Many people now use chatbots for more than simple information searches. They ask for help with work, family rows, loneliness, money worries, health questions and difficult feelings. That can make AI tools feel private and immediate, but it also means safety features need to be carefully designed.
Trusted Contact is an attempt to connect a digital warning sign with a real person. That is important because, in a crisis, a human check-in can matter more than another automated message. OpenAI says the feature is designed to sit alongside crisis helplines and emergency support, not replace them.
It also reflects a bigger issue ManyHands has covered before: chatbots can sound supportive and confident, but they are still systems with limits. In sensitive conversations, users need real-world support as well as good settings. If you have read our guide to why an overly agreeable chatbot can be a warning sign, this is part of the same wider picture.
The privacy trade-off
The privacy question is obvious. Some users may welcome the idea that someone trusted could be nudged if things became dangerous. Others may worry that a private conversation could trigger an unwanted alert.
OpenAI’s design tries to reduce that risk by making the feature optional, requiring the contact to accept, warning the user before a possible alert, and using trained human review before notification. The company also says the notification does not reveal the full chat.
Even so, no system like this can be perfect. It may miss something important. It may misunderstand a conversation. A contact may not respond well. A user may have more than one ChatGPT account, or may simply avoid the feature. Those limits are not reasons to ignore Trusted Contact, but they are reasons to treat it as one layer of support rather than a complete safety net.
What UK users should check before using it
If Trusted Contact appears in your ChatGPT settings, take a moment before adding someone. Choose a person who is likely to respond calmly, respect your privacy, and understand that an alert is a prompt to check in, not proof of exactly what happened.
It may also be worth talking to that person first. A Trusted Contact is not a therapist, emergency service or moderator. They are someone who may receive an unexpected message at a difficult moment. The clearer you both are about what that means, the more useful the feature is likely to be.
Check whether your nominated person can receive alerts in the way OpenAI supports. Check whether you can remove or change the contact later. And check the help information that explains what the contact will see if an alert is sent.
If you use ChatGPT for work, study or family admin, this is also a useful reminder to review your broader privacy settings. We have written before about ChatGPT safety labels and permissions; the same habit applies here. Sensitive settings should not be left to guesswork.
What trusted contacts should know
If someone asks you to be their Trusted Contact, treat that as a serious but manageable role. You are not being asked to monitor them constantly or solve a crisis alone. You are being asked to be reachable if the system flags a possible serious concern.
If you receive an alert, the useful response is usually simple: check in directly, be kind, avoid blame, and encourage real-world help if there is immediate risk. In the UK, urgent danger still means emergency services. For mental health crises, NHS 111, local crisis lines, a GP, or charities such as Samaritans may be relevant depending on the situation.
The important thing is not to overinterpret the notification. It is a signal to make contact, not a full explanation. The person may be embarrassed, frightened or annoyed. A calm message or call is likely to help more than demands for details.
The practical takeaway
Trusted Contact is a notable step because it recognises that AI safety is not only about what a chatbot says on screen. Sometimes the right response is to help a person reconnect with someone outside the chat.
For UK users, the best approach is thoughtful preparation. If you would want a trusted person nudged in a serious moment, consider setting it up and talking it through. If you would not, know where the setting is and review the rest of your safety and privacy controls anyway.
AI tools can be useful companions for sorting thoughts, but they should not become the only place difficult feelings go. A good safety feature should make it easier to reach people, not replace them.
