ChatGPT's 'Trusted Contact' Feature for Suicide Prevention Remains Elusive for Users

Introduction

OpenAI's ChatGPT, a powerful AI chatbot with over a million weekly users discussing sensitive topics like suicide and self-harm, has introduced an optional feature called Trusted Contact. Launched in May, this feature aims to connect individuals experiencing emotional crises and suicidal thoughts with a pre-selected contact person. However, NPR's investigation reveals that the feature is surprisingly difficult for users to find and activate, prompting concerns from mental health professionals about its accessibility and potential impact.

Key Details

Feature Purpose: The Trusted Contact feature is designed to alert a user-designated contact if ChatGPT detects serious safety risks, including suicidal ideation.
Activation: The feature is opt-in and must be manually set up by the user in the app's settings before a crisis occurs.
Discovery: NPR's own testing, along with checks with other users and a new user, found no proactive prompts or easy pathways to discover the Trusted Contact feature.
ChatGPT's Response: When directly asked about the feature, ChatGPT initially suggested the user might be thinking of another app, before acknowledging its existence and clarifying it's for serious safety concerns, not general stress.
Limitations: The feature is not available on ChatGPT's voice interface and requires advance setup; if not configured, ChatGPT cannot notify a contact even during a detected crisis.
Expert Reservations: Suicide prevention experts express mixed reactions, noting potential benefits but also significant concerns about user awareness and the necessity of proactive, non-crisis-related prompts.

Background

NPR first reported on the Trusted Contact feature in August, following a story about a young woman who died by suicide after using an older version of ChatGPT. At the time, OpenAI spokesperson Gaby Raila indicated that ChatGPT is trained to recognize risk factors and can alert a Trusted Contact if one is listed. While Raila stated the company has seen “good adoption,” no specific user numbers were provided. The intention behind the feature is to provide a safety net, mirroring the practice of listing emergency contacts in other contexts, like medical appointments.

Impact Analysis

The core issue highlighted by NPR's investigation is the lack of discoverability. Users, including experienced ones and complete novices, were unaware of the feature's existence. A test with a new user, Megan Lim, demonstrated this starkly: the chatbot welcomed her, prompted account setup and voice selection, but never mentioned Trusted Contact. Only when Megan explicitly inquired about a mental health feature did ChatGPT acknowledge Trusted Contact, describing it as an opt-in setting for serious safety concerns. This lack of proactive information means that individuals in genuine need might never set up the feature, rendering it ineffective when a crisis hits. ChatGPT's response, “If they haven't set it up, then nothing happens,” underscores this critical limitation.

“It would be ideal to offer it to all users when they are not yet in an emergency,” stated Holly Wilcox, director of the Center for Suicide Prevention at Johns Hopkins Bloomberg School of Public Health, emphasizing the need for proactive, non-emergency-based introductions to the feature.

Broader Context

The development of AI safety features, particularly for sensitive applications like mental health support, is a rapidly evolving area. While AI can offer accessible, immediate responses, the nuances of human connection remain paramount in suicide prevention. Experts like psychologist Ursula Whiteside, CEO of Now Matters Now, acknowledge that individuals struggling with suicidal thoughts sometimes turn to chatbots. Her focus groups indicated a desire for a Trusted Contact option that could be set up during clear-headed moments. However, psychologist Jennifer Lockman from the University of Alabama raises a crucial point: AI cannot replicate the empathetic, trained intervention of a mental health professional who can help individuals identify appropriate contacts and overcome internal barriers to seeking help. Lockman notes that individuals contemplating suicide often feel like a burden, and a bot might not be sufficient to break through this isolation.

Future Outlook

OpenAI spokesperson Gaby Raila stated that the company is continuously working on improving how and when ChatGPT proactively offers the Trusted Contact feature, in collaboration with psychologists and researchers. This iteration process acknowledges the challenges in balancing AI capabilities with human psychological needs. Beyond Trusted Contact, OpenAI also employs other safeguards, such as encouraging users to take breaks and providing information on crisis hotlines like 988. However, the effectiveness of these measures hinges on user engagement and awareness. For the Trusted Contact feature to fulfill its promise, OpenAI must significantly enhance its visibility and accessibility, ensuring users can set it up well before they find themselves in a critical situation. The company needs to move beyond passive acknowledgment and actively guide users toward this potentially life-saving tool.

Conclusion

While the Trusted Contact feature represents a well-intentioned effort by OpenAI to leverage AI for suicide prevention, its current implementation raises serious questions about its practical utility. The feature's obscurity means that many users who might benefit from it are likely unaware of its existence or how to activate it. Suicide prevention experts stress that human connection is irreplaceable, and while AI can be a supplementary tool, its features must be readily accessible and integrated thoughtfully into the user experience. Until OpenAI can ensure that users can easily find and set up Trusted Contact, its potential to provide a crucial safety net for those in distress remains largely unrealized.