Psychiatry ·
JAMA Psychiatry study characterizes AI psychosis in a large medical setting; model ensemble classifies mental health posts with 93.9% accuracy
Listen to this post
read by AI voices · all episodesTHE SHORT READ
A cohort study published online Oct. 7 in JAMA Psychiatry estimated the prevalence of artificial intelligence (AI) psychosis in a large medical setting and characterized affected patients. In JMIR Mental Health, an ensemble of a domain-specific classifier and a fine-tuned large language model reached 93.9% accuracy on Reddit and Twitter texts labeled by subreddit or self-stated condition, not clinical diagnosis.
THE DETAILS
- AI Oct 7, 2026A cohort study published online Oct. 7 in JAMA Psychiatry, by Zachary Bergson, Julia M. Sheffield and colleagues, estimated the prevalence of artificial intelligence (AI) psychosis in a large medical setting. According to the journal's summary, the study also characterizes affected patients and defines the role of conversational AI in documented psychotic episodes.Sources: JAMA Psychiatry, Crossref
- AI Oct 8, 2026A study published Oct. 8 in JMIR Mental Health tested models on 60,889 publicly available Reddit and Twitter texts spanning nine categories; labels came from the originating subreddit or a self-stated condition and are not clinical diagnoses. A domain-specific classifier reached 91.9% accuracy, versus 58.7% for a base large language model and 88.6% for a fine-tuned one, and a soft fusion ensemble reached 93.9%. The authors said the results are hypotheses requiring external validation on clinically characterized data before use in decision support.Source: JMIR Mental Health
The weekly innovation roundup for psychiatry, published every Sunday for the week before; every item links to its sources. Latest psychiatry news · All specialties