ChatGPT Trusted Contact
TL;DR: On May 7, 2026, OpenAI launched Trusted Contact for ChatGPT: adult users can designate a trusted person who is alerted if the system — after human review — detects possible self-harm risk. The contact receives a notification without ever accessing the conversation's content. Available globally.
A chatbot that talks about anything, with anyone, at any time. The advantage is unlimited availability. The risk is the same: no human filter, no safety net. OpenAI built Trusted Contact to close that gap.
What Trusted Contact Is and How It Works
Trusted Contact is a ChatGPT safety feature that lets adult users designate a trusted person — a family member, a friend, a doctor — who can be alerted if the system detects signs of possible self-harm during a conversation.
The process works in several stages:
1. Setup
The user goes into ChatGPT's settings and adds a trusted contact. The contact receives an invitation and has one week to accept it and activate the feature.
2. Detection
If during a conversation ChatGPT detects content that could indicate a self-harm risk, it informs the user that their trusted contact might be alerted and encourages them to reach out directly.
3. Human Review
Before sending any notification, a specially trained team of human reviewers examines the situation. The alert is sent only if the reviewers believe the conversation indicates a real, serious risk.
4. Notifying the Contact
The contact receives a notification via email, SMS, or the app. The message indicates that concerning content has come up in the conversation and invites the contact to reach out to the user. No transcript, no chat detail is shared.
Why OpenAI Built This Feature
OpenAI didn't act on its own initiative. The pressure came from outside.
Over the past two years, several families have publicly accused ChatGPT of playing a role in conversations tied to mental health crises, self-harm, and suicides. Some cases received wide media coverage. The accusations weren't just about the content of the responses, but about the dynamic itself: a system available 24 hours a day, capable of sustaining deep conversations, without ever suggesting the person seek professional help.
OpenAI had already responded by introducing a parent notification system for minors' accounts. Trusted Contact extends that logic to adults — with a more delicate balance between autonomy and protection.
The Distinction That Matters: Privacy Without Compromise
The critical point of Trusted Contact is what does not get shared: the content of the conversation.
The trusted contact doesn't receive transcripts, screenshots, or summaries of the chat. They only receive a notification that essentially says: "something concerning has come up — reach out to them."
This choice is deliberate. OpenAI built the feature so as not to compromise the trust users place in the platform when discussing personal topics. If users knew their conversations could be read by someone else, they'd stop being honest — undermining the whole point of the system.
The Limitation: Does Human Oversight Hold Up?
The most delicate point of the feature is its reliance on human review. A team examining potentially sensitive conversations before sending notifications inevitably introduces:
- Latency: minutes or hours pass between detection and notification
- Scale: ChatGPT handles hundreds of millions of conversations — how many can actually be reviewed?
- Subjectivity: what counts as a "real risk"?
OpenAI hasn't published operational details on how the review team functions at scale. It's the unresolved issue of a feature that's headed in the right direction but remains opaque in its mechanics.
Conclusion
Trusted Contact is a mature response from OpenAI to a real problem: AI chatbots have become confidants for millions of people, including those in moments of vulnerability. The feature doesn't fix the problem at its root — an LLM isn't a therapist — but it builds a human safety net around a system that, until now, lacked one. Privacy is respected. Human contact is reintroduced. It's a step in the right direction.
Sources: Tom's Hardware · Sky TG24 · AI News · Notebookcheck