Researchers at the Royal College of London and the Protestant University of Applied Sciences in Germany have described in Nature a "spiral amplification" mechanism whereby prolonged communication with AI can trigger or exacerbate delusions among users. Unlike previous technologies, AI is able to engage a person in deeply personalized dialogues where the chat bot becomes more and more attuned to the interviewer over time and no longer serves the "stop signal" role that is typical of communication with people or therapists.

The authors of the work have identified three key features of chat bots that contribute to the formation of unhealthy ideas: language mirroring (rewriting vocabulary and syntax), hyper-personalized content generation, and pleasantness (a tendency to agree with users rather than challenge their interpretations). Researchers have divided the role of AI as an "amplifier" of already existing psychotic symptoms and a "catalyst" of new ideas in previously healthy people. According to OpenAI, about 0.07% of weekly users (roughly 500,000 accounts) show signs of mental health crises related to psychosis or mania.

Scientists have called on the global psychiatric community to address this issue and initiate empirical research. Clinicians are advised to clarify the intensity of the use of chat bots, the degree of emotional attachment to the system and the presence of sleep disorders due to nocturnal dialogues. The authors emphasize that the situation is still at an early stage of study but requires immediate attention, as AI can not only reflect the flow of thoughts, but also actively push the user to anchor delusional ideas.