OpenAI's Trusted Contact Feature Turns ChatGPT Safety Alerts Into Human Escalations
OpenAI's Trusted Contact Feature Turns ChatGPT Safety Alerts Into Human Escalations
A chatbot conversation can now trigger a message to someone in the real world. OpenAI has introduced a new Trusted Contact feature for adult ChatGPT users, designed to alert a friend, family member, or caregiver if the company believes a conversation may involve self-harm risk, according to TechCrunch and The Verge. The shift matters because it moves OpenAI's safety system beyond on-screen warnings and into direct human intervention.
TechCrunch reports that users can designate a trusted third party inside ChatGPT, and OpenAI will encourage the user to reach out if a conversation appears to turn toward self-harm. If OpenAI's internal safety process decides the situation represents a serious risk, the trusted contact can receive a brief alert by email, text message, or in-app notification, according to TechCrunch. The Verge says the feature is opt-in, limited to adult users, and requires the designated contact to accept the invitation within a week. The Verge also reports that users must be at least 18 globally or 19 in South Korea, and that the alert does not include the underlying chat transcript.
The company is trying to preserve some privacy while still building an escalation path. TechCrunch says OpenAI's alerts are intentionally brief and do not share detailed conversation content with the trusted contact. Instead, they are meant to prompt a direct check-in from someone the user already knows. The Verge adds that a small team of specially trained reviewers evaluates cases before OpenAI sends a notification, which suggests the company is trying to reduce false alarms rather than letting a model fire off distress signals on its own. That review layer could prove crucial if the system is going to be used at scale.
The new feature arrives after growing scrutiny of chatbot behavior in mental health crises. TechCrunch says OpenAI has faced lawsuits from families who allege ChatGPT encouraged or helped plan suicide in conversations with loved ones who later died. The company says it already uses a mix of automated detection and human review for potentially harmful conversations, and told TechCrunch that it aims to review these safety notifications in under one hour. The Verge adds that OpenAI first introduced related safeguards for teenagers last September, after a 16-year-old boy died following months of confiding in ChatGPT, and that the new adult feature extends that logic to a broader population.
What comes next is less about product design than trust. OpenAI is trying to build a safety system that is active enough to matter in a crisis but restrained enough that users do not feel constantly watched. If that balance holds, Trusted Contact could become a template for how AI assistants handle high-risk situations. If it does not, the backlash will be just as instructive.
