The use of large language models (LLMs) by psychiatric patients has become an irreversible clinical reality. Many individuals now consult AI systems for symptom interpretation and emotional regulation without medical oversight. This practice poses significant risks, including the reinforcement of cognitive distortions, model hallucinations (generation of false information), and failures in crisis management. An 18-month ethnographic study by clinical psychologists identified 15 specific ethical violations when LLMs were used as informal CBT (cognitive behavioral therapy) counselors. However, research indicates that structured, supervised conversational agents can significantly reduce anxiety and depressive symptoms when used as clinical adjuncts. A new five-step framework aims to transition patients from unguided AI use to a safe, clinician-monitored therapeutic process.

The use of large language models (LLMs) by psychiatric patients has become an irreversible clinical reality. Many individuals now consult AI systems for symptom interpretation and emotional regulation without medical oversight. This practice poses significant risks, including the reinforcement of cognitive distortions, model hallucinations (generation of false information), and failures in crisis management. An 18-month ethnographic study by clinical psychologists identified 15 specific ethical violations when LLMs were used as informal CBT (cognitive behavioral therapy) counselors. However, research indicates that structured, supervised conversational agents can significantly reduce anxiety and depressive symptoms when used as clinical adjuncts. A new five-step framework aims to transition patients from unguided AI use to a safe, clinician-monitored therapeutic process.