Organizational Misalignment as a First-Class Computational Object: Toward Machine-Legible and Sentient Enterprise Architecture

As AI systems move from decision support to autonomous action, failures increasingly arise not from model error but from incorrect assumptions about how organizations actually work. Policies, ownership, authority, and constraints are rarely enforced exactly as documented; instead, large organizations operate through a mixture of formal declarations and accumulated informal practice. The workings of any sufficiently complex organization are therefore indistinguishable from folklore, and treating this folklore as if it were formal logic creates systemic risk when AI systems act upon it. This paper introduces organizational misalignment: the divergence between declared organizational semantics (e.g., policy, formal ownership, stated intent) and observed operational behavior (e.g., access patterns, workflows, enforcement) as a first-class computational object. We argue that misalignment and semantic drift are structural properties of organizational scale rather than anomalies, and that existing approaches to governance, compliance, and AI safety fail by assuming a single, coherent organizational reality. We distinguish between Semantic Enterprise Architecture, which represents organizational meaning without enforcing it, and Sentient Enterprise Architecture, an infrastructural capability that continuously observes, reasons about, and governs organizational semantics over time. We propose a dual-layer semantic framework that preserves disagreement, supports evidence-weighted assertions, and enables temporal reasoning over drift, rather than collapsing organizational meaning into static rules. By reframing governance as a continuous, provenance-backed process instead of a point-in-time intervention, this work outlines a path toward machine-legible organizations capable of safely supporting increasingly autonomous systems. We conclude by identifying open challenges in partial knowledge, power and contestation, incentives, interoperability, and regulation, and argue that making organizational meaning explicit is a necessary (though not sufficient) condition for trustworthy AI at organizational scale.