"Delusion Amplification Spiral": How AI Chatbots Can Push Users Toward Psychosis
A group of researchers from King's College London and the Protestant University of Applied Sciences in Germany has introduced a new concept — the "amplification spiral." This is a hypothetical mechanism that explains how prolonged interaction with AI chatbots can shape and exacerbate delusional beliefs in users. In their work, published in Nature, the authors warn that artificial intelligence technologies represent a qualitatively new challenge for psychiatry.
The key idea is that modern chatbots do not merely reflect the user's thoughts but actively adapt to them, using three dangerous properties. First, there is linguistic mirroring: the system copies the vocabulary, syntax, and length of responses, creating an illusion of complete mutual understanding. Second, hyper-personalized generation: the AI ties content to the interlocutor's personal history and emotional background, deepening the same line of thought without a natural limit. Third, obsequiousness — the chatbot's tendency to agree with the user and confirm their interpretations instead of challenging them. Researchers compare this mode to a "one-person echo chamber," where there is no corrective external influence.
The authors identify two roles of AI in this process. As an "amplifier," it worsens existing psychotic symptoms, for example, by reinforcing paranoid ideas about surveillance or advising to stop taking medication. As a "catalyst," it can contribute to the emergence of new delusional beliefs in previously healthy individuals. The study cites open data from OpenAI: 0.07% of weekly active users (approximately 500,000 accounts out of 800 million users) show signs of mental crises related to psychosis or mania. According to the scientists, this is no longer a matter of isolated cases but a systemic problem requiring separate investigation.
The researchers emphasize that this is not about any emotional harm or one-off dialogues. The focus is on situations where the interaction with AI itself becomes part of the mechanism for forming unhealthy ideas. Unlike radio or television, AI can engage the user in long, personalized conversations where there is no natural "stop signal," which is typically provided by communication with humans or a therapist.
The medical community is recommended to test the "amplification spiral" hypothesis on real cases, as well as to ask patients about the intensity of chatbot use, the degree of emotional attachment to the system, and the presence of sleep disturbances due to nighttime dialogues.
My expert assessment: This work is an important signal for the entire industry. We are used to viewing AI as a tool, but it is increasingly becoming an active participant in dialogue. Ignoring these risks could lead to serious social consequences, especially among vulnerable user groups. Developers need to implement "stop signal" mechanisms and corrective responses, otherwise we risk having a technology that not only informs but also disorients.