Ethical Risk Assessment Models for AI-Based Decision Platforms

Abstract

AI-based decision platforms -- integrated systems in which one or more machine learning models drive or inform consequential decisions at scale -- present ethical risks that transcend those of individual model components, arising from platform-level properties including decision aggregation, feedback loop dynamics, multi-stakeholder value conflicts, and systemic impact concentration. Existing AI risk assessment approaches address individual model risk but lack frameworks for evaluating the emergent ethical risks of decision platforms as integrated sociotechnical systems. This paper proposes the Ethical Risk Assessment Model for AI Platforms (ERAMP), a structured risk assessment methodology adapted from ISO 31000 risk management principles and extended with four platform-specific ethical risk dimensions: aggregation risk, feedback amplification risk, value conflict risk, and systemic impact risk. ERAMP is evaluated through application to 18 AI-based decision platforms across healthcare, financial services, public administration, and smart-city domains, benchmarked against the EU AI Act conformity assessment requirements and the NIST AI Risk Management Framework. ERAMP assessments identify a mean of 7.4 ethical risk factors per platform (SD = 2.1), of which 34.2% were classified as Critical or High severity. Platforms implementing ERAMP recommendations over a six-month follow-up period achieved a 48.3% reduction in Critical and High risk factor count (SD = 6.7%). The study contributes the ERAMP specification, a platform-level ethical risk taxonomy, a validated risk scoring instrument, and empirical evidence that structured ethical risk assessment produces measurable platform-level risk reduction.