The Authority Resolution Framework - A Five-Domain Ontology for Governing Who and What, decides at Scale.

I introduce the Authority Resolution Framework (ARF), a five-domain ontology for governing who and what holds the right to decide, at enterprise scale. Agentic AI is becoming the dominant axis of Enterprise IT spending, with industry forecasts projecting it will exceed a quarter of worldwide IT spend within a few years, even as a 2025 Forbes Research survey found fewer than one percent of executives report significant AI return on investment and Gartner has named unmanaged agentic AI proliferation, outpacing governance, its top 2026 cybersecurity trend. I argue this gap is a category-level allocation and governance problem, not a series of isolated technical failures, and that it cannot be solved by treating it as the Chief Information Officer (CIO) or Chief Digital Information Officer’s (CDIO) responsibility alone.

Enterprises maintain at least five distinct, loosely coupled ontological representations of themselves:

1. A Social Structure of Roles and Informal Influence.

2. A Business Vocabulary of Domain terms.

3. A Codified Layer of Standardized Processes.

4. A Machine-readable Layer of Executable Code and Permissions.

5. An External Real-World Context the Enterprise does not control.

Decision rights and authority are conventionally treated as a property residing within the social or policy domain alone. I argue this is a structural error with increasing practical consequence as enterprises deploy agentic AI systems capable of autonomous action; an authority relation that is not explicitly resolved across all five domains simultaneously cannot be reliably or safely delegated to a non-human actor.

I propose the Authority Relation (AR) as a formal cross-domain primitive; a six-element tuple binding an actor, action, object, bounded context, justification chain, and a novel calibration term, the DNA-Coefficient, which quantifies the divergence between an organization’s documented authority structure and its lived, practiced one. I show that the same cross-domain Object-resolution requirement underpinning the AR primitive yields semantic interoperability and consistent state-change propagation across systems as a structural consequence rather than a separately engineered concern, connecting this work to existing literature distinguishing ontologies from narrower business semantic layers.

I further argue that this same structural property has a direct, quantifiable economic benefit for organizations operating large language models at scale and I show the Authority Relation primitive transfers without modification to Physical and embodied AI, Robotics, Autonomous Vehicles, and Industrial Automation, where unresolved authority divergence carries direct physical safety consequences rather than only financial or reputational ones.

I provide a concrete implementation sketch, a JSON-LD schema, a causal-graph treatment of authority provenance, and a worked knowledge-graph query pattern to show the primitive is buildable on existing infrastructure rather than only formally specifiable, and I show this integrates directly with the Model Context Protocol and current Agentic-Memory architectures, requiring no bespoke runtime customisation.

I also propose a three-instrument empirical methodology for estimating the DNA-Coefficient (decision-log divergence analysis, organizational network analysis, and structured elicitation) and a fourth instrument specifically for auditing agent-design authority.

This work is positioned relative to foundational organizational ontology (JLG Dietz’s “Understanding and Modelling Business Processes with DEMO”; Hans Weigand, Paul Johannesson and Giancarlo Guizzardi’s “A Core Ontology of Organizational Policies”, verified directly against the open workshop precursor to the journal paper), Upper ontologies for Real-world grounding (BFO; Guizzardi et al. ‘s “Unified Foundational Ontology”), and current industry treatments of enterprise ontology for AI agent grounding, including a closely related independent three-layer ontology for neurosymbolic agent grounding (Luong Tuan & Sanyal, 2026) that addresses a different question (how an agent should reason and speak) from the one this paper addresses (whether an agent’s granted authority matches authority as actually practiced in an organisation).

I further argue, drawing a structural analogy to Weber’s account of power and “imperative control” (Herrschaft), that the probability, that a command will actually be obeyed, is as distinct from the formal right to issue it, that the deployment of agentic AI does not eliminate organizational power dynamics but relocates the primary site of leverage to whoever controls an agent’s objective function and permission boundaries. An Authority Relation that must itself be modelled, measured, and governed using the same apparatus proposed here.

I conclude with a discussion of the governance, regulatory, and audit implications for enterprises currently scaling agentic AI deployment.