LLM Psychosis: A Theoretical and Diagnostic Framework for Reality-Boundary Failures in Large Language Models

The deployment of large language models across consequential domains—clinical decision support, legal research, educational tutoring, and crisis counseling—has elevated the question of model failure modes from theoretical curiosity to practical urgency. Despite this urgency, the field's primary descriptive vocabulary fornon-trivial model errors remains centered on the term hallucination, a concept that has proven both semantically elastic and analytically imprecise (Ji et al., 2023). Broadly applied to any output that is factually incorrect, internally inconsistent, or unsupported by provided context, the hallucination label conflates mechanistically heterogeneous and risk-theoretically distinct phenomena.