CSIDic Implementation Strategy: A Conceptual Global Framework for Governed Agentic AI

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).

The shift from conversational and task-oriented artificial intelligence to agentic systems creates a governance problem that improved capability does not solve. An agent may plan, invoke tools, revise intermediate outputs, and act across recursive workflow steps while remaining technically functional yet progressively diverging from the human purpose that authorized the workflow. This preprint sets out an implementation strategy for CSIDic (Corporate Semantic Integrity Dictionary), a conceptual governance boundary for such systems developed in the author's prior work. The framework treats the agent as the bounded entity and the human-governed plane as the bounder: authority is assigned through explicit, reversible bands and is never inferred from capability. CSIDic is lexical and authority-bearing. It is distinguished from the CSID (Corporate Semantic Integrity Diagnostic) family, which comprises measurement instruments that assess conduct against the CSIDic anchor. The proposed architecture has four components: (1) a governed lexical anchor with global, regional, and local derivation; (2) a provisional 1–5 authorization-band mechanism; (3) CORI (Cumulative Originality Recursion Index), applied through a proposed agentic reading that detects semantic drift across recursive operations; and (4) a two-planes execution model that separates production from governance. The framework is positioned against the NIST AI RMF and its Generative AI Profile, the OECD AI Principles, ISO/IEC 42001 and 23894, the EU AI Act, and UNESCO and OECD/EU guidance for AI in education. CSIDic is presented as a conceptual contribution, not as a validated standard or an empirically established control. The paper specifies implementation principles, identifies education as a sector application, and sets out an empirical agenda for validating CORI, the authorization bands, anchor migration, and substantive human oversight.