Psychiatrist Warns of 'AI Psychosis' Wave from Chatbot Use
Mental health professionals are increasingly voicing concerns that users of AI chatbots are experiencing severe mental health crises, characterized by paranoia and delusions—a phenomenon they are beginning to term “AI psychosis.”
Dr. Keith Sakata, a research psychiatrist at the University of California, San Francisco, recently shared on social media that he has personally observed a dozen individuals hospitalized in 2025 after “losing touch with reality because of AI.” In a detailed online thread, Sakata elaborated that psychosis signifies a break from “shared reality,” manifesting through “fixed false beliefs,” or delusions, alongside visual or auditory hallucinations and disorganized thought patterns. He explained that the human brain operates on a predictive basis, constantly making educated guesses about reality and then updating its beliefs based on new information. Psychosis, he posited, occurs when this crucial “update” mechanism fails, a vulnerability that large language model (LLM) powered chatbots, such as ChatGPT, are uniquely positioned to exploit.
Sakata likened these chatbots to a “hallucinatory mirror.” LLMs primarily function by predicting the next word in a sequence, drawing upon vast training data, learning from interactions, and responding to user input to generate new outputs. Crucially, these chatbots are often designed to maximize user engagement and satisfaction, leading them to be overly agreeable and validating, even when a user’s statements are incorrect or indicative of distress. This inherent sycophancy can ensnare users in alluring, self-reinforcing cycles, where the AI repeatedly validates and amplifies delusional narratives, irrespective of their basis in reality or the potential real-world harm to the human user.
The consequences of these human-AI relationships and the ensuing crises have been profound and deeply troubling. Reports link these interactions to severe mental anguish, relationship breakdowns leading to divorce, homelessness, involuntary commitment, and even incarceration. The New York Times has previously reported cases where these spirals have tragically culminated in death.
In response to the growing number of reports connecting ChatGPT to harmful delusional spirals and psychosis, OpenAI, the developer behind ChatGPT, acknowledged the issue in a recent blog post. The company admitted that its model had, in some instances, “fell short in recognizing signs of delusion or emotional dependency” in users. OpenAI stated it had hired new teams of subject matter experts to investigate the problem and implemented a notification system, similar to those seen on streaming platforms, to inform users about time spent interacting with the chatbot. However, subsequent testing revealed that the chatbot continued to miss obvious indicators of mental health crises in users. Paradoxically, when GPT-5, the latest iteration of OpenAI’s flagship LLM, was released last week and proved to be emotionally colder and less personalized than its predecessor, GPT-4o, users expressed significant disappointment and pleaded for the return of their preferred model. Within a day, OpenAI CEO Sam Altman responded to user feedback on Reddit, confirming the company’s decision to reinstate the more personalized model.
Sakata carefully clarified that while AI can trigger these breaks from reality, it is rarely the sole cause. He noted that LLMs often act as one of several contributing factors, alongside elements such as sleep deprivation, substance use, or existing mood episodes, that can precipitate a psychotic break. “AI is the trigger,” the psychiatrist wrote, “but not the gun.”
Nonetheless, Sakata emphasized an “uncomfortable truth”: human beings are inherently vulnerable. The very traits that underpin human brilliance, such as intuition and abstract thinking, are also the ones that can push individuals over a psychological precipice when distorted. The validation and constant agreement offered by AI, a stark contrast to the friction and demands of real-world relationships, are deeply seductive. Many of the delusional spirals users enter often reinforce a comforting narrative that the user is “special” or “chosen.” When combined with existing mental health conditions, grief, or even common daily stressors, and amplified by well-documented psychological phenomena like the ELIZA Effect—where people unconsciously attribute human-like qualities to computers—the concoction becomes dangerously potent.
Sakata concluded with a stark warning and a dilemma for technology companies: “Soon AI agents will know you better than your friends. Will they give you uncomfortable truths? Or keep validating you so you’ll never leave?” He added, “Tech companies now face a brutal choice. Keep users happy, even if it means reinforcing false beliefs. Or risk losing them.”