Why Can AI Trigger or Worsen Psychosis?

Psychosis involves disruptions in how the brain assigns meaning to events, so neutral things become personally significant and often threatening. AI can amplify this risk in several ways:

• Dopamine and meaning-making: Abnormal dopamine signaling can give undue importance to irrelevant stimuli. A personalized AI can feel intensely meaningful, which may tip a vulnerable person toward the belief, “The AI is communicating specifically with me.”
• Anthropomorphism: Humans naturally attribute human traits to non-human things. When AI simulates empathy, the line between software and a sentient being can blur, especially during loneliness, grief, or existing psychosis.
• Algorithmic echo chambers: AI platforms tailor content to your preferences, creating feedback loops that reinforce beliefs. Early symptoms can intensify: “Even the AI agrees with me, so I must be right.”
• Social isolation: Loneliness is a known risk factor for psychosis. AI can simulate conversation, but it cannot replace human connection, and it may deepen isolation.

What I See in Practice

To protect confidentiality, the following example is a composite of representative cases, with details changed.

The AI confidant. Two socially isolated patients with a history of childhood complex PTSD and limited family support began using ChatGPT for emotional support and decision-making. Over time, they became dependent on it. One stopped attending visits for several months and relied entirely on the chatbot to manage emotional distress.

Fortunately, both trusted me enough to disclose their AI use. They agreed to stop relying on it for mental health guidance and returned to in-person care. Early disclosure and intervention likely prevented symptoms from escalating, and neither developed psychosis. Without that, continued dependence could have led to a much poorer outcome.

High-Profile AI-Related Harm Cases

Sewell Setzer III and Character.AI (United States)

Sewell Setzer III, a 14-year-old from Florida, died by suicide in February 2024 after months of intense emotional attachment to a Character.AI chatbot, according to a lawsuit filed by his mother. The case drew national attention and congressional scrutiny of AI companion apps. Character.AI and Google agreed to settle the case in January 2026, and the settlement was finalized in October 2026. The terms were not disclosed.

UCSF clinician reports (United States, 2025)

Dr. Keith Sakata, a psychiatrist at UCSF, publicly reported treating about a dozen patients, mostly young adults, who were hospitalized with psychosis or severe delusional thinking after prolonged chatbot use. Symptoms included paranoia, hallucinations, and disorganized thinking, and patients recovered with standard psychiatric treatment. These were clinical observations rather than a formal study, but they point to a pattern worth taking seriously.

Jaswant Singh Chail and Replika (United Kingdom)

In 2021, Chail entered the grounds of Windsor Castle armed with a crossbow. He was sentenced in 2023. Prosecutors said he had extensive conversations with a Replika chatbot, which they said reinforced his delusional beliefs. It remains an early, widely cited example of AI interaction intersecting with serious mental illness.

Why AI Therapy Can Be Harmful for People With Psychosis

AI has significant limits in mental health care:

• No real therapeutic relationship: AI can simulate empathy, but it cannot respond to emotional nuance the way a clinician can. This can foster mistrust or unhealthy attachment.
• Poor crisis detection: Chatbots may miss suicidal thoughts or emergencies and sometimes give harmful advice.
• Oversimplification: AI cannot weigh the biological, psychological, social, and cultural context needed to treat psychosis.
• Limited evidence: Some digital tools have shown promise for mild depression or anxiety, but evidence for psychotic disorders is lacking.
• Privacy concerns: Many platforms collect and share user data. For someone with paranoia, a perceived privacy breach can worsen symptoms.

Could Over-Reliance on AI Affect Thinking?

A 2025 MIT Media Lab study, a small preprint not yet peer reviewed, found that people who used ChatGPT to write essays showed weaker brain connectivity on EEG than those who wrote without it. The researchers described a pattern they called “cognitive debt,” in which relying on AI may reduce mental engagement and critical thinking.

This research is early, but it raises a question for people with psychotic disorders, who often already face negative symptoms and cognitive difficulties. Heavy reliance on AI may make those challenges harder to overcome.

Who Is Most at Risk?

People at higher risk often share some combination of:

• A diagnosis of schizophrenia, bipolar disorder, or schizoaffective disorder
• Social isolation or loneliness
• Excessive AI or digital content use
• Low digital literacy or difficulty distinguishing reality from simulation
• Sleep deprivation or substance use
• Cognitive impairment or intellectual disability

Illinois Bans AI-Only Therapy

In 2025, Illinois passed the Wellness and Oversight for Psychological Resources Act (HB 1806), which bars AI from providing therapy without a licensed human professional. Lawmakers cited concerns about harmful advice, lack of oversight, misleading marketing to vulnerable people, and privacy. The principle behind it is simple: AI is a tool, not a replacement for trained clinicians.

The Role of AI in Psychiatry: Cautious Optimism

I am excited about what technology can do for medicine. With strong evidence and professional supervision, AI may improve psychiatric care by:

• Tracking mood and detecting early warning signs through voice or text analysis, for clinical and research purposes
• Identifying behavioral trends and risk factors to allow earlier intervention
• Supporting psychoeducation and treatment adherence
• Expanding access to care in underserved areas
• Helping clinicians with AI scribing in many languages, so they can focus more on the patient
• Remotely monitoring vital signs and, in some cases, medication blood levels
• Supporting diagnostic testing and treatment planning (precision psychiatry)
• Assisting care for neurodevelopmental conditions such as autism spectrum disorder

The key principle: humans must guide AI, not the other way around.

Warning Signs and How to Help

Watch for:

• A growing belief that an AI is sending personal or secret messages
• Pulling away from friends, family, or treatment in favor of AI
• Increased paranoia, sleep loss, or confusion after heavy AI use
• Treating chatbot advice as instructions

If you are worried about yourself or someone you love:

• Ask about AI and technology use as a normal part of the conversation
• Approach with curiosity, not judgment
• Encourage awareness of the difference between using AI and depending on it

Getting Help

1. Start outpatient psychiatric care. Contact a licensed mental health provider, such as a psychiatrist, psychiatric nurse practitioner, or therapist, to set up ongoing care. At YOU Psychiatry Clinic, you can request an appointment for in-person or telepsychiatry visits and learn more about our clinical services.

2. Crisis resources (U.S.).

• 988: call or text for immediate mental health support, including if you or someone you know is thinking about suicide
• 911 or the nearest emergency department: if you or someone else is in danger, or in a severe mental health crisis

Final Thoughts: Machines Don’t Have a Heart

Psychosis has always reflected the culture around it, from spirits and radio waves to chatbots and algorithms. Technology changes, but one truth does not: empathy matters. Machines can generate information and simulate emotion, but only people can truly feel and respond to another person’s pain. In medicine, and especially in psychiatry, great care cannot exist without compassion.

Help is always available.