OpenAI is preparing a ChatGPT feature that could move the chatbot beyond conversation and into real-world safety intervention. Called Trusted Contact, it is designed to alert a user’s chosen contact when a chat appears to be heading toward a crisis, especially if the system detects self-harm-related language.
The feature raises a difficult question: is this a meaningful layer of protection, or an uncomfortable step into direct intervention? OpenAI appears to be betting that, in some situations, a chatbot should do more than respond inside the app and instead help connect a vulnerable user with a trusted person nearby.
How Trusted Contact works
Trusted Contact is optional and meant for adult users. Those who choose to use it can add one trusted adult contact through ChatGPT settings, with the age requirement depending on the user’s region.
The contact must be 18 or older globally, or 19 or older in South Korea. After being added, that person receives an invitation, and the selected contact still has full control to accept or decline it.
Once enabled, the feature does not require any extra button press during a chat. It runs in the background while conversations continue, which makes it a passive safety layer rather than a tool that needs to be manually triggered.
What happens when risk is detected
If ChatGPT detects a conversation that may involve self-harm, the system first warns the user that their trusted contact may receive a notification. That alert comes before anything else happens, giving the user a chance to understand that the conversation may be escalated.
OpenAI is not leaving the decision entirely to automation. The company also uses a human review team trained to assess conversations that raise safety concerns, and that team determines whether the content truly signals a safety issue.
The same review process also decides whether the trusted contact should be notified. This means the final step is treated as a safety judgment, not just a technical response from software.
Privacy is limited, not removed
OpenAI says the feature does not share the full chat with the trusted contact. The contact will not receive a transcript of the conversation.
Instead, they get a notification intended to prompt them to check on the user. That design aims to keep privacy intact while still creating a path for real-world support when needed.
This matters because chatbot conversations often contain sensitive personal details. By avoiding full message sharing, OpenAI is trying to reduce privacy risks while still making the safety feature useful.
Why the feature matters now
Trusted Contact arrives at a moment when ChatGPT is already under close scrutiny. The chatbot is widely used for nearly every kind of conversation, including highly sensitive topics, and in some cases it has been linked to self-harm situations.
OpenAI has also faced legal action related to incidents of that kind. Against that backdrop, the new feature can be seen as an attempt to strengthen protection when a conversation becomes emotionally dangerous.
Rather than relying only on in-app responses, the system is being extended toward help from people the user already trusts. That shift is likely to draw attention from both privacy advocates and safety specialists.
An added layer to existing safeguards
Trusted Contact is not the first safety measure ChatGPT has used for sensitive conversations. Before this feature, the system already encouraged users to contact help services or hotlines and suggested stepping away from ChatGPT itself.
The new option adds another layer to that approach. Its focus is not on producing a longer AI response, but on making human involvement outside the platform possible when warning signs appear.
For users who choose to activate it, Trusted Contact may serve as an additional emergency channel when a conversation begins to show signs of danger. For OpenAI, it signals a broader move toward more direct safety intervention, even if that intervention is still limited by user consent, the contact’s approval, and human review.