NPR looked for ChatGPT’s new crisis feature. It’s hard to find
With the usage and capabilities of artificial intelligence surging, people have continually found new ways to adapt it to their needs. One growing use in recent years has been in providing counsel for those struggling with self harm and suicidal ideation. According to ChatGPT, more than a million users discuss suicide and self harm with the chatbot on a weekly basis. This was a shocking number to me. OpenAi has introduced a new feature called “Trusted Contact,” that recognizes signs when someone may be at risk of self harm or suicide. In response, ChatGPT will alert their trusted contact and prompt them with steps to follow. This is an optional feature. However, neither Rhitu nor any of her colleagues had ever been prompted about this feature and they wanted to discover how people would opt into this feature. Through their testing, they found that it was buried in the settings, only works if set up in advance, and that the bot might inform them of the feature if suicidal thoughts or significant emotional distress were detected. There is still more exploration to do in this area, but there could be some promise.
OpenAI spokesperson Gaby Raila told NPR in an email that ChatGPT is trained to recognize signs that someone may be at risk — including indirect cues or warning signs that emerge over time. If the chatbot detects a serious safety risk in a user, it will alert their Trusted Contact — if they have one listed.
Curious as to how they trained the model and its accuracy. What happens if there is a false alarm?
I'm a ChatGPT user, but I'd never been alerted about Trusted Contact. Neither had my editor, nor several other users we checked with.
So, I wondered if the bot alerts new users. My colleague Megan Lim, a producer at NPR's All Things Considered, had never used the chatbot before and volunteered to give it a try.
Good to test the validity of claims.
"It only works if it's set up in advance," it added, "but ChatGPT can still respond supportively if you're feeling distressed." That support includes urging the user to reach out to a loved one or contact the National Suicide and Crisis Lifeline, 9-8-8.
OpenAI spokesperson Raila told NPR in an email that this feature isn't available on ChatGPT voice, but that the company is working on it. But when someone is typing with the bot, and the bot detects that they are struggling with suicidal thoughts and significant emotional distress, Raila said, the bot "may" tell them to set up Trusted Contact.
It’s a good start, but it seems like something more could be done. However, navigating the ethics of these types of situations would be tricky.
Psychologist Ursula Whiteside gave the feature a try. "The steps actually were quite simple and the language, and then the words were easy to follow for the most part," said Whiteside, the CEO of a suicide prevention non-profit called Now Matters Now. "Now, if I was super duper upset, it might have been harder for me to follow the process but overall I felt okay." Whiteside and her colleagues run peer support groups for individuals who are struggling with suicidal thoughts or have already survived suicide attempts, and in need of ongoing support. Many of those people now occasionally turn to chatbots in addition to their human sources of support, so they are familiar with the technology's pros and cons.
This is where it’s important to test the use-case scenario in a variety of situations. From the actual use case to those that it may not have been designed for. Testing this specific scenario could be a bit tricky with research ethics. Would have to look more into this.
Reference
Chatterjee, R. (2026, September 14). NPR looked for Chatgpt’s new crisis feature. it’s hard to find. NPR. https://www.npr.org/2026/09/14/nx-s1-5943118/chatgpt-has-a-feature-to-help-those-in-crisis-but-where-can-you-find-it
No generative AI was used in the writing of this post
