Abstract

Regulators have published high-level AI risk frameworks to guide trustworthy AI development and deployment across sectors [1]. The European Union has adopted a risk-based AI regulation that treats many AI components in cyber-physical systems as high-risk technologies requiring strong assurance [3]. Recent work on responsible AI systems emphasizes domain definition, trustworthy design and governance, underscoring the need for traceable controls across the lifecycle [5]. Clinical studies of AI-enabled healthcare show that model decisions directly affect real-world safety and quality of care in software-intensive environments [6]. Systems-theoretic safety engineering demonstrates that accidents in complex socio-technical systems often arise from inadequate control structures rather than isolated component failures [7]. Economic analyses of software automation indicate that organizations will sustain governance and reliability investments only when they deliver measurable time and cost savings [15]. This paper proposes a Unified Artificial Intelligence Governance and Reliability Engineering (AIGRE) framework that integrates governance structures, reliability and safety engineering practices, data and ML lifecycle controls, cybersecurity mechanisms, and decision-intelligence feedback loops into a single architectural view. The framework targets software-intensive and cyber-physical systems that embed learning-enabled components, providing a methodology for mapping regulatory and organizational objectives to concrete architectural decisions, lifecycle activities, and runtime indicators. Illustrative scenarios in clinical decision-support and smart infrastructure show how AIGRE can be instantiated to provide traceable links from policy objectives to operational metrics.