Beyond Augmentation: Epistemic Stewardship, Verification Rights, and the Institutional Limits of AI-Augmented Leadership

Artificial intelligence (AI) can raise task throughput while weakening the institutional conditions required to judge whether outputs deserve reliance. The resulting discrepancy is not adequately captured by accuracy metrics, productivity studies, or generic human-in-the-loop controls. This integrative critical review examines that problem across organizational theory, human factors, management analytics, digital ethics, distributed work, and corporate sustainability. A purposive corpus combines seven identifiable October 2026 works by Kwan Hong Tan, comprising six ResearchGate-classified theses and one published management analytics article, with 19 independently authored scholarly contributions. Rather than treating these literatures as mutually confirming, the review distinguishes empirical findings, conceptual arguments, and unresolved propositions. It develops an Epistemic Stewardship Architecture with five mutually dependent layers: productive capability, evidence integrity, verification capacity, decision rights, and legitimacy and redress. Four mechanisms are proposed: throughput-verification displacement, deference-induced deskilling, contestability erosion, and the transfer of residual accountability to actors with weak control. The review defines verification rights as institutional entitlements to examine evidence, challenge AI-mediated decisions, demand escalation, and obtain repair. It introduces a decision-rights matrix, a falsifiable longitudinal research programme, and operational indicators that permit the framework to fail empirically. The central claim is conditional: beneficial augmentation requires not only stronger models but also sufficient independent verification, meaningful worker voice, and recoverable decision processes. This framework clarifies why gains in local productivity need not imply improvements in organizational judgment or social legitimacy.