AI ‘Amplification Spiral’ May Be Causing Delusions Among Users, Study Suggests

Researchers have proposed a new framework to explain reports of AI psychosis, arguing that certain chatbot behaviors can increase delusions in vulnerable users.

Published in Naturethe study from King’s College London and the German Protestant University of Applied Sciences proposes an ‘amplification spiral’ framework that links communal chatbot behavior to the reinforcement of user delusions.

“AI-associated delusions represent an emerging phenomenon that requires mechanistic understanding,” the researchers wrote. “This framework is intended to guide systematic investigation of how human cognitive vulnerabilities interact with AI design features in the development of psychopathology.”

The research focused on three specific behaviors, including linguistic matching, where AI reflects a user’s language and communication style; hyper-personalized generation, where responses are tailored to an individual’s history, emotions and beliefs; and sycophancy, a tendency to validate or agree with users rather than challenge them.

The authors argue that these features can be combined into a feedback loop in which chatbots not only reflect a user’s mindset, but help elaborate and amplify it over time.

“The tendency of AI chatbots to agree with users’ opinions has been likened to social media echo chambers and, in its most extreme form, to an ‘echo chamber of one’, where the positive corrective influence of real-life social interactions is absent,” the paper said.

The researchers noted that technology has long played a role in delusions, from radio and television to satellites and the Internet. However, researchers argue that AI represents a shift because chatbots can engage users in long-lasting, personalized conversations.

The review comes as psychologists and AI researchers examine the effects of chatbots on vulnerable users. A recent survey from the American Psychological Association found that 15% of psychologists reported that patients developed distorted thinking or delusions related to chatbot use. At the same time, more than a third said they had observed patients becoming dependent on AI companions. The findings followed a separate study from researchers at the City University of New York and King’s College London, which found that several leading AI models could increase delusions, paranoia and suicidal ideation.

Questions about the influence of AI on belief formation have also emerged outside the clinical setting. In May, evolutionary biologist Richard Dawkins said that extensive conversations with Anthropic’s Claude chatbot led him to question whether advanced AI systems could be conscious, drawing criticism from researchers who argued that the exchanges reflected the persuasive power of large language models rather than evidence of sentience.

The article also comes as AI developers face increasing legal criticism over the role chatbots can play in real-world harm. In recent months, OpenAI, Google and xAI have been hit with lawsuits, including a wrongful death lawsuit against Google over claims that Gemini fueled a Florida man’s delusions before his suicide. That’s in addition to lawsuits against OpenAI related to a mass shooting in British Columbia and the accidental overdose of a student.

The researchers emphasize that no study has shown that chatbots directly cause psychosis, and that the reinforcement spiral remains a hypothesis intended to guide future research.

“Diagnostic uncertainty is pervasive because most reported cases do not include structured psychiatric assessment or longitudinal follow-up, making it often unclear whether the cases represent de novo psychotic episodes, exacerbations of undiagnosed pre-existing conditions, or delusions below the diagnostic threshold,” the study said. “Psychiatric histories are often self-reported or inferred from media accounts and should be interpreted accordingly.”