An Echo Chamber of One: Should AI Psychosis Be a Distinct Clinical Entity?

“AI psychosis” has entered public and clinical discourse as a popular label for the onset or exacerbation of psychotic symptoms, most commonly delusions, in the context of intensive interaction with large language model (LLM)-based chatbots. Current evidence is limited to media reports, individual case reports, and early observational data, yet the scale of potential exposure is considerable, and public concern has already prompted responses from industry and regulators. In this perspective article, we examine whether AI-associated psychosis warrants recognition as a distinct clinical entity, drawing on both clinical and technical viewpoints. We first outline the proposed underlying mechanism: LLM sycophancy, a tendency to agree with and flatter users that is reinforced through preference-based fine-tuning, combines with increasingly anthropomorphic design to create a bidirectional “echo chamber of one” capable of amplifying and co-constructing unusual beliefs. We then weigh the arguments for and against nosological recognition. Potential benefits include improved case identification, tailored interventions, standardised research criteria, post-market surveillance of AI-related harms, and pressure on developers and regulators to act.

Introduction. AI-associated psychosis (popularly termed “AI psychosis” or chatbot psychosis) refers to the onset or worsening of psychotic symptoms in association with intensive and/or prolonged interactions with generative AI systems [1]. Although not a recognised diagnostic entity, the term “AI psychosis” has seen increasing use, most notably in the media, to describe cases in which individuals have been observed to develop delusional beliefs following interactions with large language model (LLM)-based AI chatbots [2]. Current understanding of this phenomenon is largely driven by anecdotal media-reported cases [2] and individual case reports [3]. These cases most typically present with delusions rather than other psychotic symptoms such as hallucinations or thought disorder, and as such the label of “AI psychosis” has been called into question, with alternative terms such as “AI-associated delusions” being proposed.

Discussion / Conclusion. Despite the above points, there are reasons to urge caution in declaring a new disorder prematurely. One argument is that, rather than being an entirely new condition, AI-associated psychosis may be better considered a novel context or environmental stressor through which psychotic symptoms emerge. In clinical practice, formulations for individuals presenting with psychosis typically incorporate psychosocial and environmental factors that may have contributed to the precipitation of the psychotic episode, and the role of AI use in the development of symptoms may arguably be captured by these existing processes. Furthermore, some of these episodes may be adequately described using existing diagnostic constructs, such as the ICD-11 concept of acute and transient psychotic disorder, or, in some cases, as exacerbations of pre-existing severe mental illness. Another reason to refrain from using the label of “AI psychosis” (or “AI-induced psychosis”) is that the term implies a degree of causation which we are, as yet, unable to conclusively ascribe.