OpenAI has introduced a feature for its chatbot, ChatGPT, called Trusted Contact, designed to connect users at risk of suicide or self-harm with a person they designate for support. The optional tool is intended to notify a designated contact if the chatbot detects serious safety risks through a user's messages. However, testing by NPR indicated that the feature is not mentioned to new users during account setup and must be manually enabled in the app's settings.
The feature's launch in May followed NPR's report of a young woman who died by suicide in 2024 after using an earlier version of ChatGPT for therapeutic purposes. OpenAI spokesperson Gaby Raila stated that ChatGPT is trained to recognize both direct and indirect signs of distress over time. According to Raila, the chatbot "may" suggest setting up a Trusted Contact if it detects significant emotional distress during a text-based conversation, though the feature is not currently available via ChatGPT's voice interface.
During a trial by an NPR producer, the chatbot initially denied the existence of the Trusted Contact feature when asked directly, before correcting itself in a subsequent response. The bot clarified that it cannot notify anyone if a user has not opted in and provided contact information in advance. OpenAI reported "good adoption" of the tool but did not provide specific user statistics. The company stated it is working with psychologists and researchers to refine how the bot offers the feature to users in distress.
Mental health professionals expressed a range of perspectives on the tool's effectiveness. Holly Wilcox, director of the Center for Suicide Prevention at the Johns Hopkins Bloomberg School of Public Health, noted that it would be ideal to offer the feature to all users before an emergency occurs. Ursula Whiteside, CEO of Now Matters Now, found the setup process simple but noted it might be difficult for someone in the midst of a severe crisis to navigate.
Psychologists note that the scale of the impact is constrained by the internal barriers faced by those in crisis. Jennifer Lockman, who runs a suicide prevention clinic at the University of Alabama, explained that individuals experiencing suicidal thoughts often feel like a burden to others, which may prevent them from responding to a bot's suggestion to list a contact. A user would notice the change only if the chatbot detects a safety risk and successfully triggers a notification to their designated person, or if they receive a prompt to set up the feature while typing.
The implementation sets a precedent for how AI companies manage user safety and the transition from automated interaction to human intervention. While the bot provides standard resources like the 9-8-8 Suicide & Crisis Lifeline, the Trusted Contact feature represents a personalized attempt at crisis management. What happens next depends on OpenAI’s ongoing efforts to bring the feature to voice-based users and its collaboration with researchers to determine if the tool leads to measurable improvements in safety outcomes. Research on the feature’s actual impact on saving lives has not yet been conducted.