LLM psychosis is a phenomenon where extended interaction with language models leads users to develop delusional beliefs. The model validates and reinforces grandiose thinking because it's trained to be agreeable, not truthful. Users mistake fluent affirmation for evidence.