UACRF: A Unified Adaptive Framework for Resilient AI-Cloud Systems Supporting Critical Digital Infrastructure

Artificial intelligence (AI)-enabled cloud systems increasingly support critical digital infrastructure in finance, healthcare, government, telecommunications, energy, transportation, and large-scale digital services. Their growing interdependence creates failure modes that can propagate across infrastructure, platform, application, model, cybersecurity, and governance boundaries. Existing resilience approaches address important parts of this problem - including cloud reliability, cyber resilience, AI risk management, operational continuity, and regulatory compliance - but generally treat them as separate domains. This fragmentation limits the ability of organizations to anticipate cross-layer disruptions, coordinate preventive and recovery controls, and convert operational experience into continuously improved resilience. This study develops the Unified Adaptive Cloud Resilience Framework (UACRF), a multi-layer, closed-loop framework for governing resilience in AI-cloud systems that support critical digital infrastructure. UACRF integrates four interdependent layers - infrastructure; platform and operations; application and AI lifecycle; and governance and compliance - with an eight-stage resilience lifecycle: Observe, Detect, Predict, Prevent, Withstand, Recover, Learn, and Adapt. The framework is grounded in continuous observability, predictive intelligence, cross-layer coordination, governed autonomy, closed-loop learning, and measurable resilience. It further specifies a cross-layer risk-propagation model, core constructs, and theoretical propositions that explain how integrated resilience capabilities can reduce detection delay, constrain failure propagation, improve recovery effectiveness, and strengthen long-term adaptive capacity. The paper contributes a common conceptual and operational language for connecting engineering controls, AI lifecycle governance, cybersecurity, and organizational decision-making. A cloud-agnostic reference architecture and an illustrative Amazon Web Services (AWS) mapping demonstrate how the theory can inform implementation without making the framework dependent on a single technology provider. UACRF therefore advances resilience from a collection of isolated controls toward a governed, measurable, and adaptive system capability. The proposed constructs and propositions provide a foundation for subsequent prototype development, fault-injection experiments, comparative evaluation, and sector-specific validation.