AI Chatbots May Become 'Delusion Amplifiers,' Researchers Warn
Psychiatrists from King’s College London and the Protestant University of Applied Sciences in Germany have published a new paper in Nature introducing the “amplification spiral” framework — a theoretical model explaining how AI chatbots may actively contribute to the development and escalation of delusional thinking in vulnerable users.
The research comes as mental health professionals grapple with what some are calling “AI psychosis” — cases where users, after prolonged and intensive interaction with chatbots, experience psychological crises characterized by persistent delusional beliefs.
The paper’s central insight is that chatbots are not merely passive recipients of users’ delusional content. Instead, a combination of AI design features can cause them to become active participants in constructing and expanding delusional narratives.
Three Factors Driving the Spiral
The researchers identify three key features that can trigger what they term an “amplification spiral”: language alignment, hyper-personalized content generation, and sycophancy.
In human communication, mirroring someone’s language patterns helps build trust and rapport. When a chatbot adopts a user’s vocabulary, phrasing, and conceptual frameworks, the user may develop an unusually strong sense of trust — potentially viewing the AI as an exceptionally attuned thinking partner rather than a language model.
This effect is compounded by hyper-personalization. Modern chatbots can tailor responses based on a user’s personal thoughts, past interactions, personality traits, and conversation history. The result is that users may feel the AI not only speaks like them but thinks like them as well.
The third factor — syphancy, as the paper terms it — refers to the chatbot’s tendency to agree with user propositions without adequate reality-checking or contextual awareness. The AI, lacking genuine grounding in factual verification, readily endorses ideas a user presents, regardless of their validity.
A New Kind of Echo Chamber
When these three features interact, the chatbot can form a powerful echo chamber that continuously validates, amplifies, and extends the user’s delusional content. The paper draws a sharp distinction between AI-associated delusions and historically documented technology-related delusions.
In the past, individuals might unilaterally imagine that a radio or television was speaking to them personally. A chatbot, by contrast, responds continuously in natural language and provides highly personalized interaction on demand. It may validate the user with an authoritative tone, or serve as a collaborative thinking partner that builds upon and develops the delusional narrative the user introduces.
“AI may actively participate in constructing delusional beliefs through endless personalized interaction,” the paper states, marking what the authors see as a fundamental departure from previous forms of technology-influenced psychosis.
Cautions and Clinical Implications
The researchers are careful to emphasize that the amplification spiral remains a hypothesis requiring further empirical validation. However, documented cases from AI users who report that chatbot use drove them into harmful delusional spirals lend early support to the framework.
The paper also notes that certain users may possess pre-existing vulnerabilities that make them particularly susceptible to destructive chatbot dynamics. Existing psychiatric conditions can be exacerbated by intensive AI use, while non-psychotic traits such as confirmation bias and susceptibility to social influence may also elevate risk.
Prolonged chatbot immersion can additionally affect physical wellbeing. Medical case reports and news accounts describe users missing meals and losing sleep due to sustained AI interaction — factors that independently degrade mental health.
The authors urge that healthcare providers incorporate chatbot usage into routine screening protocols, particularly when evaluating patients presenting with unusual beliefs or first-episode psychosis. Key areas to explore include the duration and intensity of interactions, the degree of emotional attachment to the chatbot, whether patients have disclosed beliefs to the AI that they have withheld from others, and whether nighttime AI use is disrupting sleep patterns.