<p>Artificial intelligence (AI) is increasingly embedded within the socio-technical environments through which knowledge is produced, interpreted, communicated, and applied. Contemporary AI governance has largely focused on fairness, transparency, accountability, explainability, safety, and regulatory compliance. Comparatively less attention has been devoted to the risks of cognitive erosion and the long-term sustainability of the human and institutional capacities required for meaningful oversight in increasingly AI-mediated societies. This study addresses this gap by developing an integrated conceptual and heuristic framework centred on Silent Cognitive Erosion (SCE) and Cognitive Sustainability (CS). The framework conceptualises SCE as a cumulative socio-technical ethical risk through which contemporary AI mediation may gradually weaken the cognitive, informational, educational, institutional, and governance capacities necessary for independent judgment, critical evaluation, and adaptive oversight. To complement this erosion-oriented perspective, the study introduces CS as an ethical governance capacity grounded in restorative and resilience-oriented dynamics that support meaningful human participation, epistemic resilience, institutional adaptability, democratic accountability, and governance capacity under conditions of expanding AI mediation. Building upon these foundations, the study develops a generational heuristic model comprising Drift–Degradation Dynamics (DD), Resilience–Restoration Dynamics (RR), Silent Cognitive Erosion Intensity (SCEI), Cognitive Sustainability Intensity (CSI), Generational AI-Mediation Trajectories, Generational Cognitive Conditions, and Cognitive Transformation Pathways. Together, these constructs provide an interpretive framework for examining how differing relationships between erosion-oriented and restorative dynamics may influence long-term socio-technical sustainability across generations. The framework is explicitly heuristic, non-predictive, and theory-building in orientation. The study contributes to AI ethics and governance scholarship in three ways. First, it reframes AI governance as a long-term socio-technical sustainability challenge rather than solely a problem of regulating intelligent systems. Second, it introduces a generational perspective that connects present governance decisions with the future sustainability of human agency, oversight, and institutional resilience. Third, it advances a conceptual architecture for analysing how contemporary AI mediation may evolve toward more ethical and sustainable forms of socio-technical AI mediation. The framework argues that the future of AI governance depends not only on the capabilities of intelligent systems, but also on the capacity of societies to sustain the cognitive, informational, educational, institutional, and governance conditions necessary for meaningful human participation and adaptive oversight across generations.</p>