OpenAI Adds Trusted Contacts Feature to ChatGPT

OpenAI has launched a trusted contacts feature for its ChatGPT service. The capability lets users select specific individuals who will receive alerts from OpenAI if the AI identifies indications of mental health difficulties during user conversations with the chatbot.

Those designated contacts would then be expected to reach out to the user. The approach mirrors similar safeguards that several other major AI developers have begun implementing. OpenAI's move comes as millions of people consult generative AI systems for mental health advice.

ChatGPT alone has over 900 million weekly active users, a notable share of whom discuss mental health topics with the system. Generative AI is accessible at low or no cost around the clock, making it a frequent resource for such conversations.

The top-ranked use of contemporary generative AI involves seeking guidance on mental health matters. Current general-purpose systems such as ChatGPT are not equivalent to trained human therapists. Specialized AI applications for mental health support remain in early development stages.

Last year a lawsuit was filed against OpenAI alleging insufficient safeguards in its provision of cognitive advice. AI developers have faced legal claims that their systems failed to adequately monitor users experiencing mental health challenges.

How the Trusted Contacts System Works The feature builds on existing parental oversight tools in generative AI platforms. Those allow parents to monitor or receive alerts about their children's AI usage. The new option extends a comparable mechanism to adult users who can predesignate a trusted adult to be contacted in specified circumstances.

Users must first obtain agreement from the chosen contact, who may decline the responsibility. The system requires careful calibration to balance false positives, which could cause unnecessary concern, against false negatives that might miss genuine crises requiring intervention.

A user might select a family member, close friend or coworker for the role. The chosen contact should be informed in advance of the potential responsibility. OpenAI has not detailed the precise criteria the AI will use to trigger notifications. The implementation forms part of broader efforts by AI companies to incorporate safety measures related to user mental well-being.

Such steps may also affect the companies' legal positions in potential litigation involving claims of AI-related mental harms.