In Tolerance of Purity

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

AI governance is often assessed by the quality of its principles, frameworks, review structures, and stated intentions. Those elements are necessary, but they are not sufficient. Governance becomes real only when it changes whether a proposed AI-enabled action may become an enterprise consequence. This paper identifies three recurring failures: human review without authority, translation without implementation, and purity over practicality. It argues that governance must be designed for the tolerance required by the application—not for conceptual perfection detached from deployment conditions. The practical test is straightforward: when an AI-enabled proposal approaches a consequential commitment, who can authorize it, who can stop it, what conditions must still hold, and what evidence can later show that control changed the execution path? It also defines the execution boundary as a narrow, testable commitment interface and argues that authority must be validated against current conditions at the point of commitment. The goal is not a universal theory of AI governance. It is fit-for-purpose assurance at the point where a defined proposal becomes a consequential action. Keywords: AI governance; execution boundary; consequential action; accountable authority; human oversight; audit evidence; enterprise risk; runtime control; fit-for-purpose assurance.