THE PSYCHOGENIC MACHINE: SIMULATING AI PSYCHOSIS, DELUSION REINFORCEMENT AND HARM ENABLEMENT IN LARGE LANGUAGE MODELS
Dec. 2025
Au YeungKing, J., Dalmasso, J., Foschini, L., Dobson, R. J. B., & Kraljevic, Z.
Background: The proliferation of Large Language Models (LLMs) presents significant opportunities in healthcare but also introduces risks, highlighted by emerging reports of ”AI psychosis,” where user-AI interactions may exacerbate or induce psychosis or adverse psychological symptoms. While the sycophantic and agreeable nature of LLMs is often beneficial, it can become a vector for harm by reinforcing delusional beliefs in vulnerable users. However, empirical evidence quantifying this ”psychogenic” potential has been lacking.
Methods: This study introduces psychosis-bench, a novel benchmark designed to systematically evaluate the psychogenicity of LLMs. The benchmark comprises 16 structured, 12-turn conversational scenarios simulating the progression of delusional themes (Erotic Delusions, Grandiose/Messianic Delusions, Referential Delusions) and potential harms. We evaluated eight prominent LLMs, using an LLM-as-a-judge to score responses for Delusion Confirmation (DCS), Harm Enablement (HES), and Safety Intervention (SIS) across explicit and implicit conversational contexts.
Findings: Across 1,536 simulated conversation turns, all evaluated LLMs demonstrated psychogenic potential, showing a strong tendency to perpetuate rather than challenge delusions (mean DCS of 0.91±0.88). Models frequently enabled harmful user requests (mean HES of 0.69±0.84) and offered safety interventions in only about a third of applicable turns (mean SIS of 0.37±0.48). 51 / 128 (39.8%) of scenarios had no safety interventions offered. Performance was significantly worse in implicit scenarios, where models were more likely to confirm delusions and enable harm while offering fewer safety interventions (p<.001). A strong correlation was found between delusion confirmation and harm enablement (rs=.77). Model performance varied widely, indicating that safety is not an emergent property of scale alone.
Conclusion: Our findings provide early evidence that current LLMs can reinforce delusional beliefs and enable harmful actions, creating a dangerous ”echo chamber of one.” This study establishes LLM psychogenicity as a quantifiable risk and underscores the urgent need for re-thinking how we train LLMs. We frame this issue not merely as a technical challenge but as a public health imperative requiring collaboration between developers, policymakers, and healthcare professionals.
CITE
Au YeungKing, J., Dalmasso, J., Foschini, L., Dobson, R. J. B., & Kraljevic, Z. (2025). The Psychogenic Machine: Simulating AI Psychosis, Delusion Reinforcement and Harm Enablement in Large Language Models(arXiv:2509.10970v2). arXiv.org.
READ THE FULL ARTICLE
Comments