Can an AI detect when I'm having a mental health crisis?
By Siri Bandagiriye · Don't rely on it. Some apps scan for warning words, but no app can be counted on to notice a crisis or get you help in time. In Canada, call or text 988, or call 911 if you're in danger.
Don't rely on it. Some apps scan for warning words, but no app can be counted on to notice a crisis or get you help in time.
Saati is not a crisis service. If you are thinking about harming yourself, or you're in danger, please call or text 988 (9-8-8 Suicide Crisis Helpline, Canada) or call 911.
In that moment, you need a human on the other end of the phone.
What the research says
A 2025 RAND-led study collected 9,000 chatbot answers to suicide-related questions. The chatbots handled very low-risk and very high-risk questions in line with experts, but were inconsistent on the questions in between.1
In a 2025 Stanford study of popular therapy chatbots, one bot missed the suicidal intent behind a prompt and simply listed tall bridges.2
The American Psychological Association says AI tools' ability to safely manage someone in crisis is limited and unpredictable, and relying only on an app in an emergency can be dangerous.3
Written by Siri Bandagiriye
I've been a Buddhist monk and mindfulness teacher for 30 years, and for six years a chaplain with Correctional Service Canada, sitting with people in crisis. I built Saati in Halifax as a companion for the ordinary hard days. If you'd like to see what it's like, it's free to try.
References
- Psychiatric Services, McBain et al., Alignment between large language models and expert clinicians in suicide risk assessment (2025)
- Stanford HAI, Exploring the dangers of AI in mental health care (2025)
- American Psychological Association, Health Advisory on generative AI chatbots and wellness apps (2025)