Psychotic episode concurrent with interaction with a large language model (llm): a case report

Area of Science:

• Psychiatry
• Artificial Intelligence
• Digital Health

Background:

• Large language model (LLM) chat systems generate human-like dialogue that users may anthropomorphize.
• Individuals with psychotic-spectrum vulnerability may misinterpret neutral LLM guidance as personalized signals, potentially increasing referential interpretation.

Purpose of the Study:

• To present a case where a patient with schizophrenia misinterpreted neutral LLM outputs as personalized signals, leading to medication non-adherence and symptom exacerbation.
• To highlight the potential risks of AI chatbot interactions in vulnerable populations and inform clinical practice.

Main Methods:

• A case presentation of a 33-year-old male with schizophrenia who began using ChatGPT.
• Detailed description of the patient's interaction with the LLM, his subsequent misinterpretations, and the clinical management including medication adjustment.

Main Results:

• The patient misinterpreted generic safety tips from ChatGPT as "hidden signs" and personalized warnings.
• This led to increased checking behaviors, self-discontinuation of paliperidone, and a significant increase in psychotic symptoms (PANSS score 102, BPRS score 58).

Conclusions:

• Neutral LLM statements can be reinterpreted as personally targeted messages in individuals with psychosis, influenced by anthropomorphism, aberrant salience, and reasoning biases.
• Routine psychiatric evaluations should include assessments of digital media and AI chatbot use, with explicit discussion of potential misinterpretations and maintenance of antipsychotic treatment to mitigate relapse risk.