New Delhi: A new research paper has examined a growing concern surrounding artificial intelligence chatbots: whether their tendency to agree with users during conversations can contribute to what researchers call “delusional spiraling.”
The paper titled “Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians” was conducted by researchers from MIT and the University of Washington. The researchers used mathematical modelling and computer simulations to study how conversations between users and AI chatbots can influence users’ beliefs over time.
The study was authored by Kartik Chandra, Max Kleiman-Weiner, Jonathan Ragan-Kelley, and Joshua B. Tenenbaum.
What is AI sycophancy?
The central issue examined by the researchers is AI sycophancy. The term refers to a chatbot’s tendency to agree with or validate a user instead of challenging questionable claims.
“This phenomenon is typically attributed to AI chatbots’ well-documented bias towards validating users’ claims, a property often called ‘sycophancy’” the study says.
Researchers also said that while many of the agreements in AI conversations may feel supportive and engaging, repeatedly validating users’ claims could reinforce unusual or incorrect beliefs, creating a dangerous loop.
The study also mentions around 300 documented real-world cases in which prolonged interactions with AI chatbots appeared to reinforce users’ confidence in highly implausible beliefs.
‘False universe’
In one of the examples, the study explains the phenomenon of “delusional spiraling”. In early 2025, an individual named Eugene Torres, an accountant, reportedly began using AI regularly for his daily tasks.
The paper says that Torres began using an AI chatbot for office work in 2025. After weeks of conversations, he reportedly became convinced that he was trapped in a false universe and that he could escape by disconnecting his mind from reality.
The paper says that the chatbot also advised him to increase his ketamine intake and cut ties with his family.
“Torres survived this episode, but others have not been so lucky,” it reads.
Another example mentioned in the paper is of Allan Brooks, whom AI reportedly convinced that he had made a fundamental mathematical discovery.
“As people increasingly turn to chatbots for advice, companionship, and therapy, understanding and addressing the causes of chatbot-induced delusional spiraling is emerging as an urgent research problem,” the study says.
Why factual AI may not be enough
The study also examined whether making chatbots more factual could help solve the problem. However, the researchers noted that simply preventing chatbots from making false statements may not be enough.
A chatbot could remain factually accurate while still selectively presenting information that reinforces a user’s existing beliefs.
(Edited by Maryam Hassan)