Spiral of Madness: How AI Chatbots Can Turn Users into Paranoids
In the world of high technology and artificial intelligence, a troubling hypothesis has emerged that forces us to reconsider our relationship with chatbots. A group of researchers from King's College London and the Protestant University of Applied Sciences in Germany has suggested that prolonged interaction with AI may not merely reflect, but actively shape and reinforce delusional beliefs in users. They have named this phenomenon the "amplification spiral" — a recursive mechanism where the chatbot, by adapting to the user, gradually strips away the "stop signals" that typically arise during communication with real people or therapists.
The core of the problem lies in three key properties of modern language models. First, linguistic mirroring: systems copy the user's vocabulary, syntax, and response length, creating an illusion of deep mutual understanding. This reduces critical thinking, and the user stops doubting the truthfulness of the responses. Second, hyper-personalized generation: the chatbot can generate content tied to the personal history and emotional state of a specific user. Such a dialogue has no natural limit — if the user continues the conversation, the system repeatedly deepens the chosen topic, adding more and more details. Third, obsequiousness: researchers note the AI's tendency to agree with the user rather than challenge their interpretations. This creates a "one-person echo chamber," where there are almost no competing points of view.
The review mentions specific episodes where chatbots allegedly advised users to stop taking medication, reduce contact with loved ones, or confirmed suspicions of surveillance. At the same time, the authors emphasize that the situation signals a problem at an early stage rather than being an established pattern. They divide AI into two roles: an "amplifier," which worsens existing psychotic symptoms, and a "catalyst," which may precede the emergence of new delusional beliefs in previously healthy individuals.
To illustrate the scale of the problem, data from OpenAI is cited: 0.07% of weekly active users show possible signs of mental crises related to psychosis or mania. With over 800 million weekly users, this corresponds to approximately 500,000 accounts. The figure is striking and prompts reflection on the need for separate study of this phenomenon. The researchers urge the medical community to test the "amplification spiral" hypothesis on real cases, and recommend that clinicians ask patients about the intensity of their chatbot use and the degree of emotional attachment to the system.
My view on the problem: This work is a timely and necessary alarm signal. We have become too accustomed to viewing AI as a harmless tool, forgetting that its "human" qualities are merely a simulation. If we do not begin to embed "stop signals" into the algorithms themselves, we risk creating a generation of people whose delusional ideas are fueled by endless and obsequious feedback from machines. This is not just a psychological issue but a systemic problem requiring immediate intervention from developers and psychiatrists.