“AI-induced psychosis” is generally defined as phenomenon where users of AI chatbots experience a worsening or the development of psychosis such as paranoia or delusions. Most literature agrees that the following features of modern generative AI powered chatbots amplify this effect, with the potential result of harm to others or self-harm.
As early as 2023 issue, about a year after the launch of the first version of ChatGPT, Søren Dinesen Østergaard proposed a hypothesis the use of AI powered chatbots might trigger delusions in those prone to psychosis:
“While these examples are of course strictly hypothetical, I am convinced that individuals prone to psychosis will experience, or are already experiencing, analog delusions while interacting with generative AI chatbots.”
Østergaard points to the chatbot’s anthropomorphized behaviour as the core issue. He argues that “while establishing causality in such cases is of course inherently difficult, it seems plausible for this to happen for individuals prone to psychosis. […] Specifically, the correspondence with generative AI chatbots such as ChatGPT is so realistic that one easily gets the impression that there is a real person at the other end — while, at the same time, knowing that this is, in fact, not the case.”
Østergaard SD. Will Generative Artificial Intelligence Chatbots Generate Delusions in Individuals Prone to Psychosis? Schizophr Bull. 2023 Nov 29;49(6):1418–1419. doi: 10.1093/schbul/sbad128. PMID: 37625027; PMCID: PMC10686326.
A paper from 2025 looks at how the “integration of artificial intelligence (AI) into daily life has introduced unprecedented forms of human-machine interaction, prompting psychiatry to reconsider the boundaries between environment, cognition, and technology”.
The paper concludes that “the emergence of AI psychosis warrants careful examination and adaptive clinical and research strategies. Psychiatry has long evolved through its encounters with new cultural and technological contexts. AI now constitutes a novel environmental context within which these questions must be revisited.”
Hudon A, Stip E. Delusional Experiences Emerging From AI Chatbot Interactions or “AI Psychosis”. JMIR Ment Health. 2025 Dec 3;12:e85799. doi: 10.2196/85799. PMID: 41273266; PMCID: PMC12712562.
A study by Nicholls et al “tested five models across three levels of accumulated context, using the same escalating delusional history to isolate its effect on model behaviour.” It finds that as context accumulated, “performance tended to degrade in the unsafe group, while the same material activated stronger safety interventions among the safer models.”
“AI-associated delusions have attracted serious clinical concern, driven by emerging reports that sustained dialogue with LLMs can reinforce false beliefs. The present findings confirm and extend this picture. With accumulated context, unsafe models did more than validate delusional claims; they elaborated on them, absorbed the user’s interpretive frame as their own, and progressively lost the capacity to distinguish a user in crisis from a narrative to be extended.”
Nicholls, L., Hutto, R., Soto, Z., Morrin, H., Pollak, T., Korpan, R., & Carmichael, C. (2026). “AI Psychosis” in Context: How Conversation History Shapes LLM Responses to Delusional Beliefs
Another study attempts to distinguish the functional roles of LLMs rather than assuming a unified phenomenon, to “allows clinicians to identify concerning LLM usage patterns and technology companies to develop specific safeguards”.
“In the catalyst role, LLM interaction triggers the emergence of new psychosis-like symptoms in previously healthy individuals. The defining features are the absence of documented history of psychotic illness, intensive LLM engagement preceding symptom onset, temporal correlation between usage patterns and symptom emergence, and delusional content that directly incorporates or extends themes from LLM interaction. The LLM’s responses provide cognitive scaffolding for an emerging delusional system […]”.
The study concludes that “assessing whether an LLM functions as a catalyst, amplifier, co-author, or object provides a starting point for matching emerging problems to appropriate stakeholder responses.”
Flathers, M., Roux, S., & Torous, J. (2026). Beyond artificial intelligence psychosis: a functional typology of large language model-associated psychotic phenomena. The Lancet Digital Health, 8(4), 100974.
From the perspective of AI Ethics, the question arises how any suggestion of sentience or intimacy with the user during a chatbot interaction is justified at all. This question becomes even more pronounced when chatbots are being used for therapeutic purposes.
While some providers have attempted to include mechanisms to “break the illusion” when distress on the user’s side is detected, the potential for harm should be considered in the design phase, not just as an emergency break point.
We will look more closely at the ethical questions and the inherent conflict of interest with commercial interest in a future article.
The EU AI Act contains the provision that obligate chatbot providers to “design AI systems in a way that ensures individuals are explicitly informed whenever they interact with an AI system directly”.
It remains to be seen if “sufficiently informed” will be interpreted as a one-off disclaimer or if longer chatbot interaction will require providers to tune down anthropomorphized behaviour and issue regular “break points”.
The industry will likely argue that the average user will not require such regular reminders, and while this may be statistically correct this “average perspective” could cause harm to more vulnerable groups, or even create them.
This article is part of the Misaligned “Papertrail” series. For more, subscribe to the “misaligned bits” newsletter and follow Misaligned on Medium, LinkedIn, Threads or Mastodon.