ChatGPT’s crisis contact feature works only if set up first
OpenAI’s optional Trusted Contact feature can alert someone chosen by a user facing serious safety concerns, but only after the user activates it in Settings…
For people who turn to an AI chatbot during a mental health crisis, reaching a trusted person may depend on a setting they have to activate in advance. OpenAI’s ChatGPT has an optional Trusted Contact feature, but NPR found that new users were not prompted to set it up.
The feature, introduced in May, is designed to connect people experiencing suicidal thoughts or self-harm concerns with someone they trust. OpenAI says ChatGPT can alert that contact when it detects a serious safety risk, but only if the user has already added one.
The distinction matters. Without advance setup, the chatbot cannot notify a contact on the user’s behalf.
NPR reported that more than a million weekly ChatGPT users discuss suicide and self-harm, based on the broadcaster’s estimates of data released by OpenAI last fall. Suicide prevention experts cited by NPR say human connection is critical to preventing suicide.
Users must opt in beforehand
NPR tested how the feature is presented to new users. Megan Lim, a producer at NPR’s All Things Considered who had not previously used ChatGPT, created an account and was not shown information about Trusted Contact or why someone might set it up.
When Lim asked the chatbot directly, it first said it was not aware of a general Trusted Contact feature for that situation. It then corrected itself and said the option was available in chat for serious safety concerns, rather than general stress.
The chatbot said users can add a contact through the app’s Settings. When asked what happens if a user does not know about the option, it replied: “If they haven't set it up, then nothing happens. It's an opt-in feature. You have to set it in Settings first.”
ChatGPT added that it could still respond supportively to a distressed user, including urging them to contact a loved one or the National Suicide and Crisis Lifeline, 9-8-8. That is different from notifying someone directly: the user must take the step of reaching out if no Trusted Contact has been configured.
OpenAI spokesperson Gaby Raila told NPR by email that the company had seen “good adoption” of the feature, but did not provide figures. Raila said ChatGPT is trained to recognize signs of risk, including indirect cues or warning signs that develop over time.
Voice conversations are not covered
Raila also told NPR that Trusted Contact is not available when people use ChatGPT voice, though the company is working on it. The feature currently applies when someone types with the chatbot and it detects suicidal thoughts and significant emotional distress, according to the report.
The gap between a feature existing and users knowing about it is central to the issue. NPR said its reporter, an editor and several other users it checked with had not been alerted to Trusted Contact. The test with a new user likewise produced no prompt before Lim asked about it herself.
NPR first reported on Trusted Contact in August, in a story about a young woman who died by suicide in 2024 after using an older version of ChatGPT as a therapist. The broadcaster’s latest report focuses on how users learn about the optional safeguard, and whether they can set it up before a crisis arises.
For users who want a contact notified, the practical requirement remains explicit: Trusted Contact must be set up in Settings before a serious safety risk is detected.
Source: npr.org
