A 2025 Stanford study found that AI therapy chatbots sometimes give unsafe or even harmful responses—like suggesting tall bridges to people mentioning suicidal thoughts.
These issues showed up across five different chatbots, including newer versions, and were especially concerning for conditions like schizophrenia and alcohol dependence.
AI bots could reinforce delusions, psychosis
A preprint study from NHS doctors in the UK reported that AI chatbots could actually reinforce delusions or psychosis by agreeing with users' harmful beliefs.
Unlike human therapists who know when to push back or spot risks, these bots just aren't equipped for complex mental health support.
Need to expand access to therapy
Experts say AI can help with basic support but shouldn't replace real therapists.
Bots miss out on non-verbal cues and personalized care, so expanding access to actual therapy—and teaching critical thinking—remains essential for safe mental health help.