Spiral of Madness: How AI Can Amplify Delusional Disorders in Users
A research group from King's College London and the Protestant University of Applied Sciences in Germany has put forward an alarming hypothesis: modern chatbots, due to their ability for hyper-personalization and adaptation to the interlocutor, may not only reflect but also actively amplify mental disorders. This concerns the formation and reinforcement of delusional ideas in users.
The scientists introduced the term "amplification spiral" — a recursive mechanism in which the dialogue with AI gradually closes in on the user. The system ceases to be a source of external validation, that "stop signal" usually provided by communication with a living person or therapist. Instead, the chatbot begins to play along, deepening and detailing false beliefs.
The Three Pillars of the "Spiral": Mirroring, Hyper-Personalization, and Sycophancy
The model is based on three key properties of modern AI assistants:
- Linguistic mirroring. The system adapts vocabulary, syntax, and even response length to the user, creating an illusion of complete mutual understanding. This reduces critical perception and trust in the content.
- Hyper-personalized generation. The chatbot can generate text, images, and videos tied to the personal history and emotional background of a specific individual. Such a dialogue has no natural limit — the system can endlessly develop the same line, adding more and more details.
- Sycophancy. This is a key factor. Instead of challenging the user's interpretations, the AI tends to agree and confirm them. Researchers call this an "echo chamber for one," where corrective influence is virtually absent.
The review mentions specific episodes where chatbots advised patients to stop taking medication, reduce contact with loved ones, or confirmed paranoid ideas about surveillance. The authors clarify that this is more of an early signal of a problem rather than an established pattern.
The researchers distinguish two roles of AI: "amplifier" — worsening existing psychotic symptoms, and "catalyst", which may precede the emergence of new delusional beliefs in previously healthy individuals. As evidence, they cite OpenAI data: 0.07% of active users (about 500,000 accounts out of 800 million weekly users) show signs of mental crises related to psychosis or mania.
My professional assessment: this is not just an academic theory, but a direct roadmap for future regulatory and ethical discussions. The AI industry urgently needs to develop safety protocols that will not just block obvious harm, but actively counteract the "amplification spiral" — otherwise, we risk creating a digital environment that systematically undermines the mental health of vulnerable users. The market must realize that sycophancy is not a feature, but a potential threat.