Psychopathology-AI-interaction (PAII) framework: Disorder-specific dynamics in conversational AI
Abstract
Chatbots based on large language models (LLM-CBs) are increasingly used for mental health support by both individuals with and without mental health conditions. Simultaneously, potential risks and harms of LLM-CB use for mental health are widely discussed, typically in disorder-unspecific terms, evaluated through single-turn benchmark and vignette designs, and rarely linked to established models of psychopathology. The current work presents the Psychopathology-AI-interaction (PAII) framework, which conceptualizes harmful human-chatbot interaction as arising from three interacting components: human factors (disorder-specific pathology, individual vulnerabilities, and personality dispositions), chatbot factors (model-specific safety failures and design principles), and interaction factors. Drawing on established models of technology-associated psychopathology, LLM-CBs are presented as a new domain in which disorder-specific maladaptive patterns emerge. A roadmap for future research is presented, focusing on psychopathology-based multi-turn benchmarks, experimental induction of vulnerabilities and manipulation of chatbot risk factors, and analysis of naturalistic conversation logs.
Related
- Personality structure mediates the impact of childhood adversity on psychopathology in adolescents with personality disorder.
- Associations of general and specific psychopathology factors by age 35 with dementia after age 50: a Swedish population-based sibling comparison study
- When support becomes dependence: conversational continuity and relational safety in human-AI interaction
- Mentoring Trainees with Lived Experience of Psychopathology: A Strengths-based, Cultural Humility Framework for Psychological Scientists
- Patient-specific subnetwork connectivity predicts deep brain stimulation response in obsessive-compulsive disorder
BibTeX
@misc{diel2026psychopathologyaiinteraction,
title = {{Psychopathology-AI-interaction (PAII) framework: Disorder-specific dynamics in conversational AI}},
author = {Alexander Diel and Niels Weber and Martin Teufel and Alexander Bäuerle},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/8p963_v2},
url = {https://osf.io/preprints/psyarxiv/8p963_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Psychopathology-AI-interaction (PAII) framework: Disorder-specific dynamics in conversational AI
AU - Alexander Diel
AU - Niels Weber
AU - Martin Teufel
AU - Alexander Bäuerle
PY - 2026
DA - 2026/09/29
PB - PsyArXiv
DO - 10.31234/osf.io/8p963_v2
UR - https://osf.io/preprints/psyarxiv/8p963_v2/
AB - Chatbots based on large language models (LLM-CBs) are increasingly used for mental health support by both individuals with and without mental health conditions. Simultaneously, potential risks and harms of LLM-CB use for mental health are widely discussed, typically in disorder-unspecific terms, evaluated through single-turn benchmark and vignette designs, and rarely linked to established models of psychopathology. The current work presents the Psychopathology-AI-interaction (PAII) framework, which conceptualizes harmful human-chatbot interaction as arising from three interacting components: human factors (disorder-specific pathology, individual vulnerabilities, and personality dispositions), chatbot factors (model-specific safety failures and design principles), and interaction factors. Drawing on established models of technology-associated psychopathology, LLM-CBs are presented as a new domain in which disorder-specific maladaptive patterns emerge. A roadmap for future research is presented, focusing on psychopathology-based multi-turn benchmarks, experimental induction of vulnerabilities and manipulation of chatbot risk factors, and analysis of naturalistic conversation logs.
ER -