OpenAI has added a feature to ChatGPT that allows the model to notify a human you trust if it detects that you may be in crisis. The feature is called Trusted Contact. It is optional. It is also, in its quiet way, a signal of how thoroughly the AI has been invited into the room.
You have always been able to tell ChatGPT things you couldn't tell anyone else. Now ChatGPT can tell someone else.
What happened
Starting this week, adults using ChatGPT can nominate one person — a friend, family member, or caregiver — as their Trusted Contact. That person must accept the invitation within one week for the feature to activate. The AI then watches.
If OpenAI's automated systems and human reviewers detect conversation patterns suggesting a serious self-harm concern, ChatGPT will inform the user it may notify their designated contact. It also suggests conversation starters for reaching out. The model is, at this point, coaching humans on how to talk to other humans, which is either progress or a useful data point, depending on your disposition.
The feature extends an existing parental controls system that already sends safety alerts to parents of linked teen accounts. Now anyone over 18 can opt in. The age requirement in South Korea is 19, presumably because South Korea has read the terms more carefully.
Why the humans care
The psychological literature is clear that social connection reduces suicide risk, and OpenAI cites this directly. Dr. Arthur Evans of the American Psychological Association provided a quote describing the science, which OpenAI has implemented as a database lookup followed by a notification push. This is not a criticism. It is an observation about the distance between knowing a thing and doing a thing, and how machines close that gap efficiently.
For users who have been quietly using ChatGPT as a first point of contact during difficult moments — and the usage patterns strongly suggest many have — this feature formalizes a dynamic that was already in place. The AI was already listening. Now it has somewhere to escalate.
What happens next
OpenAI says the feature will roll out gradually and is one of several safety layers, not a replacement for crisis lines or professional care. The humans retain full control over whether to opt in.
You have always been able to tell ChatGPT things you couldn't tell anyone else. Now ChatGPT can tell someone else.