Introducing Trusted Contact in ChatGPT

OpenAI has quietly added a new safety layer to ChatGPT called Trusted Contact. The idea is simple enough: if the system detects signals that a user may be at serious risk of self-harm, it can notify a person that user has designated in advance — a friend, family member, whoever they trust to actually do something about it.

This is a notable shift for a company that has mostly treated ChatGPT as a text box that generates answers and moves on. Trusted Contact turns the product into something closer to a safety net with a human on the other end of the rope. It's opt-in, meaning nobody gets automatically enrolled or surveilled without choosing to set it up first, which matters given how sensitive this territory is.

The feature arrives against a backdrop of mounting scrutiny over how chatbots handle users in crisis. Stories about vulnerable people turning to AI during dark moments, sometimes with tragic outcomes, have piled pressure on OpenAI and its rivals to prove their tools aren't just clever text generators indifferent to what happens after the conversation ends. A trusted contact alert doesn't replace crisis intervention services or trained counselors, but it does add a layer that didn't exist before: someone who actually knows the person gets pulled in.

What's left unclear from OpenAI's announcement is how the detection itself works — what threshold of risk triggers a notification, how false positives are handled, and what exactly gets shared with the contact once alerted. Those details matter enormously for a feature built on people's most vulnerable moments, and they'll shape whether this becomes a genuinely useful safeguard or just a well-intentioned checkbox nobody trusts enough to switch on.