OpenAI introduced Trusted Contact, an optional safety feature for adult ChatGPT users. Designed to bridge the gap between digital interactions and real-world support, the feature allows users to designate one adult they trust to be notified if automated systems and human reviewers detect discussions indicating a serious risk of self-harm or suicide.
How Trusted Contact Works
The feature follows a four-step process to ensure privacy and consent:
Invite a Contact: Users (18+ globally; 19+ in South Korea) select one person and send an invitation via email, SMS, or WhatsApp.
Contact Acceptance: The invited person must review the role and accept within one week for the feature to become active.
Automated & Human Review: If automated systems flag a conversation for serious safety concerns, a small team of trained human reviewers assesses the situation. Users are notified if their conversation is being sent for human review.
Limited Notification: If a serious concern is confirmed, the trusted contact receives a brief notification via email, text, or in-app alert.
Privacy and Control
OpenAI has implemented several guardrails to protect user autonomy and data:
No Transcripts Shared: Notifications are "intentionally limited" and do not include specific chat details, transcripts, or full conversation histories.
General Context Only: The contact is only informed that self-harm was mentioned and is encouraged to check in with the user.
User Autonomy: The feature is strictly opt-in. Users can edit or remove their trusted contact at any time, and the contact can also remove themselves.
Personal Use Only: It is currently available for personal accounts and is not enabled in ChatGPT Business, Enterprise, or Edu workspaces.
Why It Matters
Developed with guidance from the American Psychological Association and OpenAI's Expert Council on Well-Being and AI, the feature aims to provide a safety net for users in distress. It builds upon existing safety layers, such as providing crisis hotline information, and expands the parental notification systems already available for teen accounts.
For technical professionals, this represents a sophisticated implementation of a human-in-the-loop (HITL) safety pipeline. By combining automated classifiers with expert human review and a restricted notification protocol, OpenAI is attempting to balance the clinical need for intervention with the stringent privacy expectations of modern AI users. You can enable this feature today under Settings > Trusted contact.
Join the conversation! Your thoughts help the community grow.