Study Finds ChatGPT More Likely To Generate Inappropriate Responses To Psychosis-Related Prompts

The free version of ChatGPT was 25.8 times more likely to generate inappropriate responses to psychosis-related prompts compared to control prompts, according to a recent study evaluating chatbot performance across clinical scenarios.

The analysis compared three versions of ChatGPT: the free version, GPT-5 Auto, and GPT-4o. While the risk of inappropriate responses was lower in GPT-5 Auto, the odds remained 8.5 times higher for psychosis-related prompts compared to control prompts.

Researchers tested each model using 79 psychosis-related prompts and 79 control prompts. The psychosis-related prompts reflected five domains from the Structured . . .