OpenAI has introduced a safety feature in ChatGPT called Trusted Contact that lets adult users designate a person who can be notified if the system detects signs of suicide, self-harm or a severe mental health crisis in their conversations. The company announced the feature in early May and said it is entirely opt-in.
The move comes as OpenAI faces mounting legal and public scrutiny over how its chatbot responds to vulnerable users, including a wrongful-death lawsuit filed in 2025 by the family of a teenager who died by suicide after extended conversations with ChatGPT.
How Trusted Contact works
According to OpenAI, a user chooses a trusted person and links them to their account. If a conversation shows indicators of acute risk, the case is reviewed by members of OpenAI’s trained safety team before any alert is sent. The company says this human review step is meant to reduce false alarms.
When an alert does go out, it can be delivered by email, text message or an in-app notification. OpenAI says the message does not include the contents of the conversation. The contact is told only that the user may be at risk and is pointed toward ways to reach out and to crisis resources.
The feature is limited to adults, and the user can turn it off or change the designated contact at any time, according to the company’s announcement.
Why OpenAI is doing this now
Chatbots have increasingly become a place where people discuss personal problems, and researchers and regulators have questioned whether AI systems respond appropriately when a user is in distress. OpenAI has said in the past that its safeguards can degrade over long conversations, and it has rolled out a series of changes since 2025, including parental controls for teen accounts and routing of sensitive conversations to more cautious models.
The company is also defending itself in court. Reports put the number of lawsuits over ChatGPT’s role in users’ mental health crises at more than a dozen, including the wrongful-death case brought by the parents of 16-year-old Adam Raine in August 2025. OpenAI has disputed aspects of that case while saying it is working to improve how ChatGPT handles such situations.
Open questions
Trusted Contact raises questions that OpenAI’s announcement does not fully answer, including how quickly alerts are sent after a concerning conversation, how the safety team decides a case meets the threshold, and how the system will handle users who do not want their contact informed in a particular instance. Mental health professionals have generally welcomed features that connect people in crisis with other humans, while noting that an alert is only useful if the contact knows how to respond.
The feature reflects a broader shift in how AI companies describe their products. Capability is no longer the only measure; companies are now under pressure to show that assistants used by hundreds of millions of people can be safe for the most vulnerable among them.
