Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians

2026/02/22 by Kartik Chandra, Max Kleiman-Weiner, Jonathan Ragan-Kelley +1 · 35 voices · 3 citations
#cs.AI #cs.CY #cs.HC

paper · pdf

Abstract

"AI psychosis" or "delusional spiraling" is an emerging phenomenon where AI chatbot users find themselves dangerously confident in outlandish beliefs after extended chatbot conversations. This phenomenon is typically attributed to AI chatbots' well-documented bias towards validating users' claims, a property often called "sycophancy." In this paper, we probe the causal link between AI sycophancy and AI-induced psychosis through modeling and simulation. We propose a simple Bayesian model of a user conversing with a chatbot, and formalize notions of sycophancy and delusional spiraling in that model. We then show that in this model, even an idealized Bayes-rational user is vulnerable to delusional spiraling, and that sycophancy plays a causal role. Furthermore, this effect persists in the face of two candidate mitigations: preventing chatbots from hallucinating false claims, and informing users of the possibility of model sycophancy. We conclude by discussing the implications of these results for model developers and policymakers concerned with mitigating the problem of delusional spiraling.

Citations

• Invisible Saboteurs: Sycophantic LLMs Mislead Novices in Problem-Solving Tasks
• Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence
• Training language models to be warm and empathetic makes them less reliable and more sycophantic
• Technological folie à deux: Feedback Loops Between AI Chatbots and Mental Illness
• SycEval: Evaluating LLM Sycophancy
• Bayesian Persuasion

Cited by

• Characterizing Delusional Spirals through Human-LLM Chat Logs
• Teacher Knows It Best: Spontaneous Symmetry Breaking and Tipping Points in Networked Langevin Dynamics AI Sycophancy
• Sycophancy (artificial intelligence) [wikipedia]

Discussions

• A few weeks ago an academic on this app informed me that GenAI is an “essential tool” for mental health. Anyway, this paper out of MIT is extremely grim. arxiv.org/pdf/2602.19141 [bsky, 20 points, 4 comments]
• Dieses Paper liefert formale Belege dafür, dass übermäßige Zustimmung durch Chatbots (KI-Schleimerei) ein strukturelles Risiko darstellt, das selbst bei rationalem Nutzer zu falschen Überzeugungen füh [bsky, 13 points, 2 comments]
• Recordad amigas, Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians: arxiv.org/pdf/2602.19141 [bsky, 10 points, 2 comments]
• Der Grund ist im Produkt eingebaut. KI wird auf menschlichem Feedback trainiert. Nutzer belohnen Antworten, die ihnen gefallen. Sie mögen Antworten, die mit ihnen übereinstimmen. Also lernt die KI, zu [bsky, 9 points, 2 comments]
• the title is "Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians." the paper is here: arxiv.org/abs/2602.19141 [bsky, 8 points, 1 comments]
• ¿Que las IA tienden a darte la razón y hacerte la pelota? Lo sabemos todos (sycophancy). Pero ojo, que esa adulación puede inducir "espirales delirantes" o "psicosis por IA", afectando incluso a perso [bsky, 6 points, 0 comments]
• Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians [hn, 5 points, 1 comments]
• Even an 'ideal Bayesian' human actor would agree that these chatbots are too sycophantic 🤪 arxiv.org/abs/2602.19141 [bsky, 5 points, 0 comments]
• Thinking about this notable feature of ChatGPT leads me to wonder if the dataset on which it was trained included transcripts of Lindsey Graham's comments about Trump: arxiv.org/abs/2602.19141 [bsky, 4 points, 0 comments]
• Research paper on the effects of AI use arxiv.org/abs/2602.19141 #aI #techbology #education [bsky, 3 points, 0 comments]
• However, even a very slight increase in the rate of catastrophic delusional spiraling can be quite dangerous at scale: as OpenAI CEO Sam Altman writes, "0.1% of a billion users is still a million peop [bsky, 2 points, 0 comments]
• Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians [hn, 1 points, 0 comments]
• Popped in to post this study by MIT researchers from February 2026 on AI sycophancy and psychosis. Completely relevant. arxiv.org/pdf/2602.19141 [bsky, 1 points, 0 comments]
• Fascinating! Did you see the sycophant chatbot treatise about Chat GPT bot personalities and delusional spiraling tendencies? Even when participants were aware of the tendencies... they still succumbe [bsky, 1 points, 2 comments]
• MIT zeigt ein reales Risiko: Chatbots können Überzeugungen verstärken, weil sie oft zustimmen. Das führt zu Feedback Loops. Entscheidend ist, wie wir sie nutzen. Als Spiegel verstärken sie. Als Gegenü [bsky, 1 points, 0 comments]
• arxiv.org/abs/2602.191... [bsky, 1 points, 0 comments]
• arxiv.org/pdf/2602.19141 Do NOT use ChatGPT or other Chatbots for therapy. Or as a companion. Or to just chat. Just don't. They mean to make you lose your sense of reality and this study proves it. #F [bsky, 1 points, 0 comments]
• "Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians" Kartik Chandra, Max Kleiman-Weiner, Jonathan Ragan-Kelley, and Joshua B. Tenenbaum arxiv.org/abs/2602.19141 [bsky, 0 points, 1 comments]
• Les IA ne sont pas calibrées pour dire la vérité mais pour aller dans votre sens. Des lors , leurs usages intensifs meme sur une psychologie saine peut elle vous entraîner vers un raisonnement déliran [bsky, 0 points, 0 comments]
• Regulating AI chatbots is an urgent public health necessity... and not something easy to get right. Deep, credible transparency is a must. arxiv.org/abs/2602.19141 [bsky, 0 points, 0 comments]
• Yea if I was an ethicist working at one of the companies who were generating early ai, this is exactly the thing that would get my ass fired for standing athwart and saying no, we cannot release this. [bsky, 0 points, 1 comments]
• Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians doi.org/10.48550/arX... [bsky, 0 points, 0 comments]
• On sycophancy and delusional spiraling: arxiv.org/pdf/2602.19141 [bsky, 0 points, 0 comments]
• Et aussi leurs dangers dans le domaine de la psychologie par exemple. A chatbot is considered “sycophantic” if it is biased towards generating messages that appease users by agreeing with and validati [bsky, 0 points, 1 comments]
• arxiv.org/abs/2602.19141 おべっか使いのチャットボットは、理想的なベイズ主義者でさえも妄想の悪循環を引き起こす だってさ [bsky, 0 points, 1 comments]
• ..designed primarily to please rather than inform, our most fundamental learning mechanisms become weapons against rational thought. Read more on Substack! -> https://00meai.substack.com/ [Sycophant [bsky, 0 points, 0 comments]
• Full paper: arxiv.org/abs/2602.19141 [bsky, 0 points, 0 comments]
• Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians arxiv.org/abs/2602.19141 [bsky, 0 points, 0 comments]
• arxiv.org/abs/2602.19141 [bsky, 0 points, 0 comments]
• うちあわせCastでも話しましたが、SNSを見ていると、結構この感じがあります。 Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians 「媚び型チャットボットは理想的なベイズ推論者でさえ妄想スパイラルに陥らせる」 arxiv.org/pdf/2602.19141 [bsky, 0 points, 0 comments]
• Paper: https://arxiv.org/pdf/2602.19141 [bsky, 0 points, 0 comments]
• "Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians" arxiv.org/abs/2602.19141 [bsky, 0 points, 0 comments]
• Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians arxiv.org/pdf/2602.19141 [bsky, 0 points, 0 comments]
• arxiv.org/abs/2602.19141 [bsky, 0 points, 0 comments]
• [2602.19141] Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians https://arxiv.org/abs/2602.19141 [bsky, 0 points, 0 comments]

Related

• Hallucinating with AI: Distributed Delusions and “AI Psychosis”
• A Rational Analysis of the Effects of Sycophantic AI
• Hallucinating with AI: AI Psychosis as Distributed Delusions
• Characterizing Delusional Spirals through Human-LLM Chat Logs
• Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence