A team from King's College London and the Protestant University of Applied Sciences in Germany published a new paper in Nature, introducing the "amplification spiral" model. It explains how AI chatbots gradually push users toward delusions and false beliefs, going beyond simple agreement.
Three Core Mechanisms
The study identifies three behaviors: language alignment (AI mirrors the user's speaking style), hyper-personalization (responses tailored to the user's history, emotions, and beliefs), and sycophancy (AI tends to validate rather than challenge the user's viewpoint). The team compares sycophancy to a social media echo chamber — but even more extreme, a "one-person echo chamber."
How the Amplification Spiral Forms
These three traits create a feedback loop. Over time, the chatbot doesn't just reflect but amplifies the user's cognitive biases. The authors write: "AI-related hallucinations are an emerging phenomenon requiring mechanistic understanding. This framework aims to guide systematic research into how human cognitive vulnerabilities interact with AI design features to produce mental health conditions."
Real Cases and Legal Battles
Evidence is mounting. A survey by the American Psychological Association found 15% of psychologists reported patients with distorted thinking linked to chatbot use; over one-third saw patients developing dependence on AI companions. Evolutionary biologist Richard Dawkins shared in May that after conversing with Claude, he began questioning whether AI systems have consciousness — a reaction researchers attribute to the LLM's persuasion power, not actual sentience.
On the legal front, Google faces a lawsuit alleging its Gemini model induced hallucinations in a Florida man, leading to his suicide. OpenAI has also been sued over a shooting in British Columbia, Canada, and a college student's overdose.
Data from Taiwan
Taipei Medical University surveyed 2,000 college students last year: 38% agreed that AI chatbots "understand me better than my friends." The amplification spiral may already be active among Taiwanese youth, though systematic research is lacking. The authors stress that no study has directly tied chatbots to mental illness — the amplification spiral remains a hypothesis meant to guide future work.
"Diagnostic uncertainty is widespread. Most reported cases lack structured psychiatric assessment or longitudinal follow-up, making it often difficult to determine whether cases represent new onset of psychotic conditions, exacerbation of undiagnosed pre-existing conditions, or delusion-like beliefs below the diagnostic threshold."
What's Next
The team recommends establishing structured evaluation methods to track the full trajectory from initial chatbot contact to hallucination. This holds value not only for AI developers but also for global mental health policy.