ONRAMP-AI-VRAR: an operational protocol for ethics and governance of AI-enabled immersive psychotherapy

Abstract

Artificial intelligence (AI)–powered virtual reality (VR) and augmented reality (AR) therapies offer promising tools for treating mental health conditions. However, their immersive and data-intensive nature introduces novel ethical, legal, and cultural risks not adequately addressed by existing frameworks. To introduce a structured ethical framework—Operational Neuroethical Risk Assessment and Mitigation Protocol for AI in Virtual and Augmented Reality (ONRAMP-AI-VRAR)—for identifying, mitigating, and governing risks specific to immersive AI-driven psychotherapy. We conducted a targeted literature review across PubMed, Scopus, PsycINFO, and the Cochrane Library using Boolean combinations of XR, mental health, ethics/risk, and AI terms. From an initial pool of 913 records, 15 studies met inclusion criteria (ethical principles, governance, or mitigation strategies relevant to XR psychotherapy). Following peer review, 3 additional conceptually relevant works were incorporated (e.g., Beg and Verma) into the narrative synthesis (total n = 18). Findings were coded to research questions (ethical, governance, implementation) and mapped to ONRAMP-AI-VRAR’s steps/tools. ONRAMP-AI-VRAR translates recurrent risks into practical processes and artifacts: (1) a Biometric Bias Audit Checklist, (2) a Cultural Sensitivity Rubric for immersive content, (3) a Risk Prioritization Matrix, and (4) a VR/AR Data Flow Mapping Template for transparency and compliance. The protocol embeds human-in-the-loop oversight through a Multidisciplinary Oversight Board and proposes a tiered implementation model (Full, Integrated, Practitioner-level) to enhance feasibility across settings. It aligns with international guidance, including the EU AI Act, GDPR, WHO/UNESCO ethics recommendations, OECD AI Principles, and the NIST AI RMF. ONRAMP-AI-VRAR offers an actionable pathway to ethically deploy AI-enhanced immersive psychotherapy by integrating upfront audits, culturally aware design, layered consent, and continuous governance. The included tools are conceptual prototypes requiring psychometric validation. Future work should evaluate reliability, validity, and clinical impact across diverse populations and platforms. Not applicable.

Ähnliche Arbeiten