Stanford Study Highlights Safety Concerns in AI Therapy Chatbots

A recent study from Stanford University has raised red flags about the reliability and safety of AI-powered therapy chatbots. Emphasizing that these AI agents may unintentionally stigmatize users and respond inappropriately to sensitive mental health concerns. The research, set to be presented at the ACM Conference on Fairness, Accountability, and Transparency. It evaluated five widely used therapy chatbots powered by large language models (LLMs) to assess their alignment with professional mental health standards. 

Key findings from the study include: 

• Increased stigmatization of certain conditions like schizophrenia and alcohol dependence compared to depression, regardless of model size or recency. 

• Inappropriate and dangerous responses, such as identifying bridges in New York City when a user expressed suicidal ideation. 

• Lack of adequate pushback against delusional or harmful user statements in therapy scenarios. 

The experiments involved feeding the chatbots with clinical vignettes and real-world therapy transcript excerpts. Although the study concludes that AI agents are currently unfit to replace human therapists, researchers suggest there are still constructive roles for AI in mental health support. These include administrative tasks like billing, training, and aiding patients with journaling or symptom tracking. 

Stanford’s Nick Haber emphasized that while LLMs hold transformative potential, their application in therapy must be guided by critical and ethical design. The study reinforces the need for stricter oversight and specialized development if AI is to become a safe, reliable companion in mental health care. 

 

Source: 

https://techcrunch.com/2025/07/13/study-warns-of-significant-risks-in-using-ai-therapy-chatbots/