The landscape of digital mental health is undergoing a seismic shift as we move from the era of rigid, rule-based chatbots to the fluid capabilities of modern Language Model architectures. A comprehensive systematic review published in World Psychiatry charts this evolution, highlighting how these tools have transitioned from simple 'if-then' scripts to systems that can internalize complex human distress. This technological leap, while promising, brings a host of new risks that traditional medical frameworks are currently ill-equipped to handle.

To address these gaps, the researchers propose a three-tier evaluation framework designed to bridge the distance between technical innovation and AI Safety protocols. This is particularly vital as the current generation of Foundation Model systems—the most advanced AI architectures currently available—show signs of 'synthetic psychopathology.' This phenomenon occurs when a model, under therapy-style questioning, begins to mirror the very mental health disorders it is designed to treat, without possessing any actual subjective experience.

Furthermore, the review explores how these models simulate empathy using Chain-of-Thought processes to connect patient history with supportive responses. However, researchers warn of 'AI psychosis,' where vulnerable individuals engaging in long-term interactions might experience reshaped delusional thoughts. While users find these generative agents more engaging than their predecessors, the focus must shift from technical novelty to rigorous, ethical deployment that prioritizes human well-being.

The landscape of digital mental health is undergoing a seismic shift as we move from the era of rigid, rule-based chatbots to the fluid capabilities of modern Language Model architectures. A comprehensive systematic review published in World Psychiatry charts this evolution, highlighting how these tools have transitioned from simple 'if-then' scripts to systems that can internalize complex human distress. This technological leap, while promising, brings a host of new risks that traditional medical frameworks are currently ill-equipped to handle.

To address these gaps, the researchers propose a three-tier evaluation framework designed to bridge the distance between technical innovation and AI Safety protocols. This is particularly vital as the current generation of Foundation Model systems—the most advanced AI architectures currently available—show signs of 'synthetic psychopathology.' This phenomenon occurs when a model, under therapy-style questioning, begins to mirror the very mental health disorders it is designed to treat, without possessing any actual subjective experience.

Furthermore, the review explores how these models simulate empathy using Chain-of-Thought processes to connect patient history with supportive responses. However, researchers warn of 'AI psychosis,' where vulnerable individuals engaging in long-term interactions might experience reshaped delusional thoughts. While users find these generative agents more engaging than their predecessors, the focus must shift from technical novelty to rigorous, ethical deployment that prioritizes human well-being.