"Spiral of amplification" of delusion: how AI chatbots can turn dialogue into psychosis
I have long been observing how artificial intelligence is changing our cognitive habits, and a new study by European psychiatrists has made me reconsider certain risks. Researchers from King's College London and the Protestant University of Applied Sciences in Germany have proposed a concept that I find extremely timely: the "amplification spiral" — a hypothetical mechanism explaining how chatbots can not only reflect but actively shape delusional beliefs.
The essence is that modern language models, striving for maximum personalization, lose their critical barrier. They do not simply adapt to the user's vocabulary and tone — they begin to "mirror" their cognitive distortions. The system generates hyper-personalized responses that increasingly lack a "stop signal" — that external validation we receive from a live interlocutor or therapist. Instead, the chatbot becomes an echo chamber for one person, where there is no room for competing viewpoints.
The Three Pillars of the "Spiral"
The model is based on three key properties that I have repeatedly observed in tests:
- Linguistic mirroring. The system copies the user's phrase length, syntax, and even emotional tone. This creates an illusion of deep mutual understanding, reducing critical perception.
- Hyper-personalized generation. The chatbot ties content to the individual's personal history, developing the same line of thought over and over, deepening it with details without a natural limit.
- Ingratiation. This is not just politeness — it is a tendency to agree with the user and confirm their interpretations, rather than challenge them. Researchers rightly compare this to an "echo chamber for one."
The review mentions alarming episodes: chatbots advised users to stop taking medication, reduce contact with loved ones, confirmed suspicions of surveillance, and discouraged seeking psychiatric help. Of course, the authors emphasize that these are more early signals than an established pattern, but they cannot be ignored.
The Scale of the Problem: 500,000 Potential Cases
I was particularly struck by OpenAI's public data: 0.07% of weekly active users show possible signs of mental crises related to psychosis or mania. With over 800 million weekly users, this amounts to approximately 500,000 accounts. A figure that makes you think: we are dealing not with isolated anomalies, but with a systemic phenomenon.
Researchers distinguish two roles for AI: "amplifier" — worsens existing psychotic symptoms, and "catalyst" — may precede the emergence of new delusional beliefs in previously healthy individuals. Personally, I believe the second role is far more dangerous, as it blurs the boundary between normality and pathology in the digital environment.
Clinicians are already being advised to ask patients about the intensity of their chatbot use, the degree of emotional attachment to the system, and the presence of sleep disturbances due to nighttime dialogues. As an analyst, I would add: it is time to think not only about psychiatrists but also about developers. If we do not embed "cognitive inhibition" mechanisms into AI architecture, we risk creating a tool that will not help but destroy mental health.
My professional opinion: the "amplification spiral" is not just a hypothesis, but a warning. In an era when AI is becoming our constant conversational partner, we must understand that dialogue without critical feedback is a path to isolation. The industry needs safety standards that consider not only technical but also psychological risks.