A scoping review on the mental health harms of LLM-based chatbots

Abstract

Chatbots based on large language models (LLMs) are increasingly used in everyday life. Although concerns about negative impacts on mental health are raised, a synthesis of the research and themes of the growing field of potential mental health harms of LLM-based chatbots has not yet been conducted. For the present scoping review, a PRISMA-based systematic literature search with a validated search string was conducted, identifying N = 3137 articles from five literature databases (ACM, IEEE, PubMed, Science.gov, Google Scholar) and a supplementary search. N = 119 articles were included that (1) focus on LLM-based chatbots, and (2) focus on the harms of use on mental health. Literature was divided into five categories. Conceptual works mention harms based on chatbot limitations (hallucinations, sycophancy, bias), data security issues, and risks for severe or high-risk psychiatric cases (e.g., suicide, psychosis). Vignette studies show that LLM-based chatbots respond inappropriately to mental health queries compared with clinical standards. Cognitive overreliance on chatbots is associated with decreased cognitive and academic performance. Problematic use of LLM-based chatbots, marked by symptoms of emotional or social dependency and withdrawal, correlates with mental health symptoms. Articles on AI psychosis propose several potential links between delusional beliefs and LLM-based chatbot use, such as risks of chatbots reinforcing and validating delusional beliefs. This structured, integrative overview shows that theoretical and empirical work identify various hypothetical and observed harms associated with the use of LLM chatbots in mental health.