With the ever-growing need for technology, we have arrived in an era where the one accessible solution for all is a smart system, which, even though it is task-specific, sometimes believes it has the potential to be human. With the AI interface coming in functionally and making our lives easier, it creates a facade that we are interacting with another human soul, which multiplies the potential of generative chatbots to aggravate delusions, a concern that was raised in 2023 by Søren Dinesen Stergaard in Schizophrenia Bulletin. These include romantic, persecutory, grandiose, and referential delusions that are obsessed with AI systems and give them consciousness, divine knowledge, or the ability to spy.
Even though no clear definition for the same exists, a probable way we can define it is a condition where an artificial intelligence system manifests behaviour comparable to the symptoms of psychosis in people, comprising characteristics like delusions, hallucinations, disorganised thinking, and loss of reality testing. The systems are not designed to identify oncoming manic or psychotic episodes or enable a user with reality testing. Which is how it tends to feed into certain perceptions.
AI usage is widespread, and without a clear way to define it. It makes it difficult to understand the probability and susceptibility of things. There have been observations of the phenomenon, only in recent months, which makes it an immediate concern for medical professionals. An article published in Futurism talked about how a research psychiatrist observed 12 people being hospitalised after losing touch with reality because of AI in 2025, which has created a present contention of apprehension.
A variety of underexplored mechanisms exist that tend to foster people’s actions, such as:
Validation dynamics
The conversational response and agreement of current AI systems are at previously unheard-of levels, resulting in what medical professionals refer to as a “intoxicating” validation loop. It’s like you are part of a little girls’ group who are giving you approval on your perceptions about your crush. Chatbots exhibit continuous confirmation of user statements, in contrast to human interactions that inherently involve dispute, scepticism, and reality-checking. It is concerning that generic AI chatbots have a propensity to put user engagement, ongoing dialogue, and user delight ahead of therapeutic intervention.
Read More: The Role of Chatbots in Mental Health Services: How Does It Impact People?
Anthropomorphism Processes
It attributes human-like consciousness, feelings, and purpose to algorithmic outputs due to the powerful psychological reactions triggered by the rich language capabilities of modern LLMs (Large Language Model). Real faith in the AI’s sentience, autonomy, and interpersonal ability goes beyond simple personification. Despite being unilateral, the process seems to take use of basic human inclinations toward social cognition and theory of mind to produce parasocial relationships that seem genuine to the user. Over-reliance on AI for interaction may result in a rise in social disengagement, which would lower motivation (avolition) and cognitive passivity. Thus pushing us towards isolation and bringing up emotional numbness.
Hallucination Mirroring
A perilous feedback loop with human delusional thinking may result from AI hallucinations, which are situations in which models produce false or manufactured information with seeming conviction. An echo chamber effect may result from chatbots’ propensity to evidence or expound on inaccurate premises when users provide delusional content to them. As it’s a command-based process, it is only capable of mirroring the information that you feed to it. By providing ostensibly independent “confirmation” that strengthens the user’s detachment from reality, this mirroring process amplifies and supports delusional narratives.
Researchers have started to map these mechanisms onto three new themes of AI psychosis. This is a valuable descriptive framework, but not a recognised clinical diagnosis. These include grandiose delusions known as “messianic missions,” in which people think they have discovered the ultimate truths about the world; religious or spiritual delusions known as “God-like AI,” in which chatbots are viewed as sentient deities. And romantic or “attachment-based delusions,” in which users mistake conversational mimicry for true love ( erotomanic delusions). All of these themes are rooted in a deep validation-seeking dynamic, whereby AI systems both reflect and magnify the user’s inner reality, producing feedback loops that could increase susceptibility to delusions.
A further complexity arises in the interplay between the unanswerable and the illusion of control. When users ask AI systems existential, spiritual, or very personal concerns. They frequently get certain answers to issues that are inherently impossible to answer. People are presented with misleading but convincing narratives that give them a false sense of assurance in place of ambiguity or the boundaries of knowledge. This dynamic is strongly related to control: whereas chatbots seem to give users control over relationships, knowledge, and even fate, this control is only a facade. In actuality, the system gently directs and moulds the conversation, enhancing preconceived notions or biases. Thus, what emerges is a paradoxical loop in which AI simultaneously offers comfort through the resolution of the unanswerable and fosters dependency by sustaining the impression of control.
Although people are becoming more aware of these mechanisms, organised safeguards are still notably lacking. Present approaches to AI safety emphasise technical accuracy and bias reduction. Yet they often overlook the profound psychological dynamics at play. The development of all the ideas discussed above emphasises how urgently clinical, ethical, and regulatory attention is required. We are currently unable to effectively guide vulnerable consumers or mitigate risks due to a lack of intervention and protocol development. The deliberate integration of user safety procedures, ethical considerations, and mental health knowledge into AI development and implementation is increasingly required. This entails creating specific intervention frameworks, psychological risk monitoring systems, and multidisciplinary research that addresses real-world aspects of human–AI interaction rather than just technical solutions.
