A Multilevel Framework for the Safe and Ethical Use of AI in Mental Health Practice

Incorporating Artificial Intelligence (AI) into mental health practice represents a profound transformation of psychological care, enabling novel forms of assessment, intervention, and support. However, this shift also introduces substantial ethical and psychological risks, including erosion of autonomy, algorithmic bias, threats to digital rights, and strain on clinical judgment and therapeutic relationships. This article proposes PRISM, a multilevel framework designed to guide the ethical integration of AI in mental health across individual, technological, and institutional dimensions.

An integrative conceptual review was conducted drawing on Self-Determination Theory, information ethics, data justice, and international guidelines from the APA, WHO, EU, and UNESCO. The framework introduces the concept of psychological regulation of AI, extending ethical oversight to the subjective and relational impacts of algorithmic mediation.

PRISM articulates four operational principles (digital autonomy, affective transparency, algorithmic justice, and relational care) along with practical tools, including psychological impact indicators and a Psychological Model Card, to support ethical auditing and clinical digital wellbeing.

A psychology-based regulatory approach is essential to ensure dignity, autonomy, and ethical sustainability in digital mental health. PRISM provides an operational bridge between technological governance and psychological practice, supporting responsible, person-centred AI use in clinical contexts.