Human-Centered Artificial Intelligence in Mental Health Nursing: Clinical Augmentation, Therapeutic Relationships, and Responsible Governance

Artificial intelligence (AI) is entering mental health nursing through prediction, natural language processing, digital phenotyping, conversational agents, and generative documentation. This integrative narrative review examines how these functions can support care while preserving therapeutic relationships, recovery, professional accountability, and patient rights. A targeted literature and regulatory update was completed on October 26, 2026; the revised bibliography comprises 35 sources, including six contemporary nursing or nursing-education reviews. Sources were synthesized thematically, without pooled effects or a claim of exhaustive systematic retrieval. Recent reviews establish substantial conceptual overlap and identify persistent limitations in external validity, person-centred outcomes, sustained engagement, and implementation evidence. Accordingly, this article contributes an author-developed nurse-led implementation and assurance framework that connects relational nursing theory, sociotechnical safety, evidence maturity, intended-use risk, and Brazilian governance. Operational requirements include calibration, prevalence-sensitive predictive values, prospective silent validation, measurable escalation, competent human review, abstention, version control, incident grading, and controlled revalidation. Documentation benefits must be assessed through total verification workload and protected therapeutic time. The risk matrix is a conceptual proposal, not a validated clinical instrument or binding legal classification. The framework has not been prospectively evaluated. AI adoption is defensible only when local evidence, service capacity, equity, privacy, and patient choice support the intended use; technical performance alone cannot establish clinical benefit.