Currently, artificial intelligence (AI) is clinically relevant to mood and anxiety care, but the evidence base is uneven across use cases.This naartive review synthesizes recent literature most relevant to clinicians and investigators.Five themes dominate the current field: patient-facing adjunctive tools, failure modes and safety risks, clinician-facing decision support, passive sensing and measurement infrastructure, and governance.Recent randomized evidence supports a narrow efficacy claim for structured chatbot interventions, with small improvements in depressive and anxiety symptoms and more J o u r n a l P r e -p r o o f consistent effects on engagement than on symptom superiority.These studies do not support autonomous psychotherapy, and they do not establish a therapeutic advantage for open-ended large language model systems over more constrained designs.Safety studies, by contrast, identify active concerns: harmful endorsement, weak youth risk assessment, inconsistent crisis handling, and anxiety/OCD reassurance loops.The strongest current clinical signal lies in supervised clinician-facing decision support, where recent trials of AIassisted antidepressant selection improved treatment persistence and some downstream symptom outcomes.Passive sensing detects behaviorally meaningful signals, but evidence that alert-driven deployment improves care remains insufficient for routine practice.Across stakeholder and policy sources, the most defensible deployment model is human-inthe-loop, stepped, and bounded by explicit handoff rules.
Martin P. Paulus (Fri,) studied this question.