The Psychosis Risk Interaction Scoring (PRIS) Protocol: A Framework for Detecting and Mitigating AI-Induced Psychosis Risk in Conversational Agent

The rapid integration of large language models (LLMs) and conversational agents into everyday life has created unprecedented opportunities for human–AI interaction, alongside emerging concerns regarding psychological safety. While existing moderation frameworks address hate speech, misinformation, and toxicity, no standardized protocol currently exists to detect and mitigate AI outputs that may increase psychosis risk in vulnerable users. This paper introduces the Psychosis Risk Interaction Scoring (PRIS) Protocol, a six-domain evaluative framework designed to identify, score, and intervene in AI-generated content that could contribute to psychosis-related symptoms such as delusions, paranoia, hallucinations, and disorganized thinking. PRIS assigns risk scores on a 0–18 scale across six domains—Reality Distortion, Paranoid Reinforcement, Grandiosity Induction, Thought Insertion, Hallucination Encouragement, and Disorganized Logic—supported by a structured evaluation pipeline that integrates content analysis, contextual review, and optional user-state data. The protocol’s design allows seamless mapping to the Synthetic Consciousness Assessment Battery (SCAB), enhancing its applicability in AI governance systems. Ethical considerations, implementation strategies, and potential applications in AI safety, clinical screening, and computational psychiatry are discussed. PRIS offers an interdisciplinary bridge between mental health research, natural language processing, and AI safety engineering, establishing a foundation for evidence-based governance of conversational AI systems.