AI chatbots for teens may miss suicide risk in subtle messages, review warns

A review says current chatbot tests often use one-off messages, and performance drops with indirect, unclear, or slowly shared risk.

Editorial illustration — not from the study.

In this paper in npj Digital Medicine, the authors review research on how general-purpose chatbots and AI “companions” detect and respond to suicide risk disclosures from young people. They report that most available studies rely heavily on single-turn, researcher-written prompts that are not designed with youth communication in mind.

The review also finds that chatbot performance drops when risk is shared indirectly, ambiguously, or over time—patterns common in adolescent communication. The authors say the real-world impact of chatbot interactions on youth suicide risk remains unknown and outline a clinical research agenda focused on youth input, better risk-detection evaluation, and safer connections to crisis support.

How current tests fall short

The review found that existing studies of AI chatbots and youth suicide-risk disclosures use almost entirely single-turn, researcher-generated prompts rather than testing approaches designed for youth. It reports that chatbot performance degrades substantially with indirect, ambiguous, or gradually disclosed risk. The authors also state that the real-world impact of chatbot interactions on youth suicide risk has not been established.

Why safety evidence is missing

Because many youth are using generative AI chatbots for mental health support, failure to recognize or handle suicide risk messages correctly could matter for how quickly a crisis response is triggered. The authors emphasize that these safety questions are not answered by the current research base and call for a structured agenda to test safer ways to detect risk and connect youth to trusted adults and crisis resources.

Evidence base and key caveats

This is a narrative review of existing research on chatbot responses to suicide-risk disclosures, and it assesses how studies are designed and how performance changes under different message styles. A key limitation is that the review highlights that current evidence is based largely on artificial, one-turn prompts, leaving real-world effects on youth suicide risk unknown. The paper calls for prospective outcome research to fill that gap and notes that chatbot capabilities are changing rapidly.

AI food images matched real food on health and calories — but not appeal

People judged AI-generated food images similarly to real food for healthiness and calorie content, but found them less realistic and were less willing to eat them. The results suggest AI images may be useful for controlled food-perception studies, but they are not interchangeable with images of real food.

20 hours ago·3 views

Society & Economics

Preprint finds measurable patterns in how AI skills emerge and fade

A preprint reports measurable patterns for how capabilities emerge and are forgotten during language-model training. The study says emergence depends on several model parts becoming useful together, while forgetting depends on how widely and strongly a capability is stored.

2 days ago·5 views

Society & Economics

In nursing mice, brain cells link bone, fat, energy use and pain sensitivity

A study in nursing mice identified hypothalamic nerve cells that linked prolactin signaling with bone and fat metabolism, food demand and pain sensitivity. Reducing parts of this pathway lowered food demand and pain sensitivity, but the findings have not been established in humans.

3 days ago·12 views