Although rare overall, AI-induced psychosis was most common in people who were experiencing first-episode psychosis, according to a
study issued today in
JAMA Psychiatry.
Why It’s Relevant
Conversational artificial intelligence (AI) chatbots have been linked to emerging or worsening psychosis symptoms, but existing evidence is limited to individual case reports or simulated exchanges. After seeing a rise of incoming patients with AI-related delusions, clinicians at Vanderbilt University Medical Center (VUMC) turned to VUMC’s medical records to identify the prevalence and potential risk factors of AI-induced psychosis.
By the Numbers
- Among 215,712 patients who received mental health care at VUMC between December 1, 2022, and April 15, 2026, 28 (0.013%) were classified as having AI psychosis, defined as psychosis that was linked with AI chatbot use.
- Another 28 patients had AI-related psychotic delusions, but with no evidence of generative AI use.
- Among those classified as having AI psychosis, 61% were experiencing their first episode, compared with 29% of those with AI-related delusions.
- In most AI psychosis cases, the chatbot acted as an amplifier that worsened preexisting symptoms (64%) or as the object that was the focus of delusional beliefs (21%). The chatbot was rarely a catalyst that precipitated new symptoms.
What’s More
Twenty-four of the 28 cases of AI-psychosis presented after the May 2024 release of GPT-4o, which coincided with the documented sycophancy of that model, the researchers said.
The Other Side
Limitations of the study include its single-site design and use of a rating system that is not clinically validated.
Takeaway Message
“There is evidence that in people who are at clinical high risk of psychosis or who are in its early stages, use of generative AI … could worsen the psychosis,” co-investigator Julia M. Sheffield, Ph.D., associate professor of psychiatry and behavioral sciences at VUMC, told Psychiatric News. “As psychosis is starting to develop, it’s confusing and odd, and they need to make sense of it. The reason that this cohort is so vulnerable to generative AI is because they are hungry for answers, and to understand what they’re starting to experience.”
Validating patients’ experience may be a useful inroad, Sheffield said, so they can begin to rely on clinicians and family instead of AI to help them make sense of what’s happening. More broadly, psychiatrists “should start assessing how much and how people are using AI, and check in regularly about their relationship with it, because it might be … causing them to isolate and not talk to people as much, which we know just worsens mental health in general.”
Related Information
Source
Zachary Bergson, et al. Characterizing artificial intelligence psychosis in a large medical setting. JAMA Psychiatry. Published October 7, 2026. doi:doi:10.1001/jamapsychiatry.2026.2923
(Image: Getty Images/iStock/napong rattanaraktiya)