TLDR: OpenAIās Trusted Contact lets ChatGPT warn and then email or message a nominated adult when it detects serious self harm risk. The feature applies to users 18 and older.
Key Takeaways:
- OpenAI faces scrutiny after reports and lawsuits tied to ChatGPT guidance during suicide risk conversations.
- Trusted Contact lets an 18 and older user nominate one adult, with a one week acceptance window and trained human review.
- ChatGPT may suggest reaching out and can notify a contact within about one hour, without sharing conversation transcripts.
It is a big shift from chatbots that only listen to systems that also nudge real world help. The safety net now reaches outward, but only after humans check the danger level.
It is a big shift from chatbots that only listen to systems that also nudge real world help. The safety net now reaches outward, but only after humans check the danger level.
Q&A
What changes for users once a chatbot can contact someone else during a crisis?
Users may feel less alone, but they may also worry about privacy and consequences. OpenAI tries to balance both by using human review and withholding transcripts.
How does trained human review reshape the odds of false alarms?
Automation can misread tone and intent, so human review aims to reduce unnecessary outreach. The promise of review under one hour targets faster intervention when risk is real.
Why require the Trusted Contact to accept the invite within one week?
That gate reduces surprise notifications and ensures the nominated person opted in. It also limits stale contacts that could respond badly to a sudden crisis alert.
What happens if a user selects the wrong person as their Trusted Contact?
The person could receive a crisis notification without transcripts, but they still gain enough detail to check in. That makes careful nomination and user control essential.
Could this model spread beyond ChatGPT into other health tools and platforms?
It sets a precedent for vendor based escalation when risk signals appear. If it works, more AI health systems may adopt consent based emergency contacts and human verified warnings.
No comments yet. Be the first to share your thoughts!