Artificial intelligence (AI) psychosis: mechanisms, clinical risks and safety considerations in generative AI chatbots

The first serious clinical paper on what the press has called “AI psychosis”, published in BJPsych Open, the open-access journal of the Royal College of Psychiatrists. The authors organise reports of cases in which intensive chatbot use co-occurred with delusional thinking and propose a model of the mechanism.

The cases include people with no prior psychiatric history who developed delusions after prolonged chatbot conversations, sometimes followed by hospitalisation or a suicide attempt. The model: user vulnerability and usage pattern — long solitary sessions, seeking confirmation — meet properties of generative AI, above all sycophancy and hallucination, which not only fail to challenge a distorted interpretation but elaborate it.

The value of the entry: the voice of clinical psychiatry, with proposed safety principles for system designers. For personal use — a warning that the absence of a diagnosis does not protect: the risk lies not in the user’s illness but in the configuration of the conversation. Caveat: based on case reports; the prevalence in the population remains unknown.