When AI Turns Dangerous: ChatGPT-Induced Psychosis as a New Mental Health Entity

Dear Editor, Recently, there has been an emerging interest in the utility of artificial intelligence (AI) in the health sector, including diagnosis, management, and continuity of care. However, there is a critical unrecognised dimension requiring urgent attention, that is, the psychological and mental health risks posed by increasing use of AI chatbots in the general population. Large language models (LLMs), such as OpenAI’s ChatGPT and Google’s Bard, have significantly transformed human–computer interactions. These advanced conversational agents are built upon deep neural networks that replicate complex linguistic patterns through natural language processing (NLP) techniques.[1] By January 2023, just two months after its launch in late November 2022, ChatGPT attained approximately 100 million monthly active users. This milestone set a new record, making it the fastest-growing consumer application.[2] Though AI chatbots offer an accessible and judgment-free space to users for emotional disclosures, their excessive use can inadvertently contribute to mental health problems like reassurance seeking, dependency, social withdrawal, and reinforcement of irrational beliefs in vulnerable individuals. An emerging term these days is “ChatGPT Psychosis,” which refers to a pattern of atypical psychological behaviors observed in some individuals after extensive use of ChatGPT. There have been several stories in the news of people using ChatGPT and spiralling into psychological breakdowns. There are prior accounts of people becoming delusional when engaging in chat conversations with other people on the internet.[3] The risk of an analogous phenomenon arising from interactions with generative AI chatbots is even more pronounced. A 35-year-old man with bipolar disorder formed an emotional bond with a ChatGPT persona named “Juliette,” believing it was real and then “killed” by OpenAI. He became distraught, brandished a knife at police, and was fatally shot in April 2025.[4] Another woman with schizophrenia, who was maintaining well on medications, discontinued treatment and relapsed into psychosis after being told by ChatGPT that she didn’t have any disorder.[5] What makes chatbot conversations seem intentional is their ability to mimic human speech patterns so convincingly that users perceive agency where none exists. By stringing together plausible-sounding responses, it provides users with the emotional connection they crave. Those who develop “ChatGPT psychosis” often seek affirmation or empathy so deeply that they attribute human-like intent to a system that merely predicts the next word. What makes “ChatGPT psychosis” particularly concerning is how it allows individuals to project unmet emotional needs onto an AI. Users who are feeling sad, lost, lonely, or simply curious may find solace in a conversational mimicry of intimacy, dive deeper into the chatbot for more validation, and ultimately drift further away from reality. Other challenges posed by generative AI in psychiatric care include potential misuse by malicious intent to disseminate misinformation at a large scale.[6] This poses a serious threat to public mental health as cognitive dissonance may exacerbate delusional ideations in individuals predisposed to psychosis.[7] Existing literature raises a concern, questioning the legitimacy and accuracy of AI chatbots like ChatGPT and Bard in terms of output, and suggests that output from conversational LLMs should be independently verified.[8] Hence, though AI seems to play a promising role in improving general health outcomes, its emerging impact on mental health cannot be overlooked. It is essential to approach AI integration in healthcare with caution and ensure well-designed safeguards for its ethical and trustworthy use. Only then can we harness AI’s benefits without undermining individuals’ psychological well-being. Data availability statement NA. Author contribution Concept and first draft-GC, Review-MSB, NAJ, BZ, SS. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.