Domain-Specific Constitutional AI: Enhancing Safety in LLM-Powered Mental Health Chatbots

This project advances the safety of mental health chatbots by adapting Constitutional AI (CAI) with domain-specific principles tailored to psychological care. General AI safeguards often fail to address the unique risks of mental health applications, including crisis escalation, therapeutic guideline adherence, and sensitive dialogue handling. Our framework trains large language models with mental health–specific safety rules, enabling more reliable crisis intervention, reduced misinformation, and improved trust in emotionally vulnerable contexts. By enhancing both accuracy and safety, this work lays the foundation for scalable, trustworthy LLM-powered tools in therapy, crisis detection, and wellness support.

Publications:


Full paper

Project information

  • Category: Future Healthcare, Large Language Models, and Personal Health Models
  • Contact Person: Amir M. Rahmani
  • Domain-Specific Constitutional AI: Enhancing Safety in LLM-Powered Mental Health Chatbots

info@futurehealth.uci.edu

© Copyright 2021 UCI Institute for Future Health - All Rights Reserved