OpenAI today announced a new safety feature for adult ChatGPT users focused on mental health.
Starting today, ChatGPT users can add what the company calls “reliable connection” Who can be notified if AI’s automated systems and trained reviewers determine that a user is engaging in discussions about self-harm.
The new feature comes amid growing research into the impact of artificial intelligence and other digital platforms on mental health. Last year, OpenAI revealed that 0.07% of its weekly users reported “mental health emergencies involving psychosis or mania”, 0.15% were at risk of “self-harm or suicide”, and another 0.15% showed signs of “emotional reliance on AI”. Given the company’s claims, about 10% of the world’s population uses ChatGPT weekly, which is roughly three million people.
The trusted connection feature extends ChatGPT’s existing parental safety notifications, which alert parents when a linked teen’s account shows signs of concern. Presented by Instagram similar parental warnings earlier this year.
OpenAI now provides these warnings to its adult users. The company said the feature was developed with guidance from mental health and suicide prevention clinicians, researchers and organizations.
“Trusted Contact is designed to encourage contact with someone the user already trusts,” the company said in its announcement. “It is not a substitute for professional care or crisis services and is one of several layers of provision to support people in difficult situations.”
OpenAI added that ChatGPT will still encourage users to contact crisis hotlines or emergency services when necessary.
The feature can be activated by any user 18 years of age or older through ChatGPT’s settings. From there, users can designate another adult to serve as a trusted contact by providing details such as the contact’s phone number and email address.
A trusted contact will then receive an invitation explaining the feature and have one week to accept. If they decline, the original user can assign another connection instead.
Once the feature is enabled, OpenAI’s automated monitoring systems may note that a user may be discussing self-harm in a way that raises serious safety concerns. The system will then inform the user that a trusted contact may be notified and encourage them to contact them directly. It will even provide some recommended conversation starters.
A small team of specially trained reviewers will then assess the situation and determine whether reporting to a trusted contact is appropriate, the company said.
If OpenAI decides to send an alert, a trusted contact can receive it via email, text message, or in-app notification. The alert will only explain the general reason for reporting self-harm and prompt a trusted contact to sign up. It will also include instructions for navigating those conversations.
OpenAI noted that the notifications will not include specific details or chat transcripts to protect user privacy.
