Ethics-Aware Software Engineering Frameworks for AI Applications

Abstract

The integration of artificial intelligence (AI) into software-intensive systems demands that software engineering (SE) practice evolve to encompass ethical properties including fairness, transparency, explainability, and harm avoidance. Despite widespread scholarly consensus, existing SE frameworks offer limited operational guidance for embedding ethical requirements throughout the software development lifecycle (SDLC). This study proposes and empirically evaluates an Ethics-Aware Software Engineering (EASE) framework extending established SE practices with ethics-specific processes, artefacts, and verification criteria. The EASE framework is evaluated through a controlled study involving 18 professional SE teams (n = 162 engineers) from six European organisations developing AI-enabled products across healthcare, finance, and smart-city domains. Teams adopting EASE achieved a 41.3% reduction in ethics-related defects, a 29.6% improvement in stakeholder fairness satisfaction, and a 22.4% reduction in time-to-ethics-review per sprint. The study contributes a replicable framework, a validated ethics-requirements taxonomy, and empirical benchmarks for AI software engineering practice and regulatory compliance.