PaperlayerView exact revision on arXiv

Source-linked AI summary

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

Parviz Shariff

arXiv:2608.15832v1cs.AI

TL;DR

AI agents need authority resolution, not merely technical capability, before consequential actions. This paper proposes ARF, a five-domain ontology whose Authority Relation represents authority provenance, scope, and contextual validity across enterprise domains.

  • Problem

    Existing ontologies can represent policy specifications without validating whether documented authority still holds in practice.

  • Method

    ARF models authority as a relation resolved consistently across Social, Business, Process, Machine, and Real-world domains.

  • Results

    The framework defines a cross-domain Authority Relation and adds practice-sensitive validation to ontology-based authority representation.

  • Takeaways & Limitations

    Authority resolution is positioned as a knowledge-representation problem requiring cross-domain consistency and attention to authority as exercised.

  • Takeaways & Limitations

    The Authority Relation currently represents one Object per relation and does not capture partial, multi-step obligations.

Abstract

from arXiv · show

As AI systems become increasingly capable of autonomous action, determining whether an agent is technically capable of performing an action is insufficient: the system must also determine whether the action is authorised in its context. This paper introduces the Authority Resolution Framework (ARF), a five-domain ontology for representing and resolving authority across organisational roles and informal influence, business concepts, codified processes, machine-readable permissions and executable systems, and external real-world context. ARF defines the Authority Relation (AR) as a cross-domain primitive binding an actor, action, object, bounded context, justification chain, and a calibration measure termed the DNA-Coefficient, which captures divergence between documented authority structures and authority as practiced. The framework provides a machine-interpretable representation of authority provenance and scope, with JSON-LD representations and knowledge-graph query patterns for authority resolution. ARF is designed to support AI agents in determining the provenance, scope and contextual validity of authority before executing consequential actions. The framework positions authority resolution as a knowledge-representation and reasoning problem at the intersection of ontology engineering, semantic AI, agentic AI and AI governance.

1. Introduction

The paper argues that agentic AI requires more than technical capability or shared business meaning: systems need a shared, traceable account of who may do what across enterprise representation layers. It proposes the Authority Resolution Framework (ARF), which resolves authority as a cross-domain relation and measures divergence between formal authority and authority in practice through the DNA-Coefficient.

  • 1. Introduction: Agentic AI deployments fail to scale partly because enterprises lack a shared, traceable account of who is permitted to do what across representational layers.The paper distinguishes this governance problem from model capability and data architecture alone.
  • 1. Introduction: Authority and decision rights are commonly represented separately in organisational charts, RACI matrices, access-management permissions, or organisational ontologies, but none alone resolves the full problem.These representations are individually necessary yet insufficient when treated as isolated or simply combined.
  • 1. Introduction: ARF represents one underlying authority relation through five domains: social, business, process, machine, and real-world facts.The domains cover actual influence, operational meaning, codified procedures, system enforcement, and effects in the world.
  • 1. Introduction: When the five representations drift across teams and timelines, governance may exist on paper without being enforceable in practice, a gap agentic AI exposes by reducing compensating human judgment.The paper frames cross-domain resolution as a response to this recurring enterprise pathology.
  • 1. Introduction: ARF treats authority as a relation that must resolve across domains and introduces the DNA-Coefficient to capture divergence between documented authority structures and authority as exercised.The framework also connects cross-domain object resolution with preserving meaning as entity state changes propagate between business, process, machine, and real-world representations.
  • 1. Introduction: The paper contributes a six-element Authority Relation primitive spanning five enterprise ontology domains, alongside procedural co-ownership requirements and a 6-level Authority Maturity Model.The contributions require independent business, technical, and governance affirmation before a DNA-Coefficient score is considered valid.

2. Related work and Positioning

ARF positions itself as a cross-domain authority-resolution layer that builds on organizational, upper-ontology, industry, and governance research without competing with those traditions. Its distinctive contribution is checking whether documented authority still holds in practice, while acknowledging limits to formalizing human judgment and culture.

  • Organizational ontology: Organizational ontology grounds ARF’s Social Domain, especially patterns formalizing organizational policies, authority delegation, and the distinction between policy documents and interpreting IT systems.ARF adopts these patterns for an Authority Relation’s documented half rather than competing with them.
  • Organizational ontology: ARF’s current single-Object Authority Relation cannot represent partial, multi-step obligations, which the paper identifies as an open extension.The limitation is contrasted with policy-pattern treatments of obligations and authority provenance.
  • Organizational ontology: An ontology that represents what a policy specifies does not establish whether that specification still holds in practice; ARF adds that validation layer.This distinction motivates ARF’s DNA-Coefficient as a scored, directional, per-instance measure rather than a general claim about policy effectiveness.
  • Upper ontology: BFO and UFO provide domain-independent grounding for ARF’s Real-world Domain, which ARF uses without extending their upper-ontology apparatus.The Authority Relation’s Object must resolve consistently against the enterprise’s chosen upper ontology.
  • Industry practice and explainability research: Luong Tuan and Sanyal’s three-layer enterprise-agent ontology was validated through 1,800 runs, whereas ARF focuses on verifying whether delegated authority remains real.The paper presents the approaches as complementary: one supports agent reasoning and speaking, while ARF resolves authority validity.
  • Theoretical grounding: ARF’s theoretical grounding treats authority as only partially formalizable, using non-contradiction, dual-process judgment, and shared structural variation to motivate calibrated authority resolution.Kahneman’s distinction specifically supports treating the DNA-Coefficient as an adjustment rather than a complete formalization of trust or culture.

3. Formal Framework

The Formal Framework defines five enterprise ontology domains and an Authority Relation as a six-element cross-domain primitive. It uses this structure to represent authority scope, provenance, delegated agency, and divergence between documented authority and practice.

  • 3.1 The Five Enterprise Ontology Domains: ARF enumerates five named enterprise ontology domains as scaffolding for resolving authority relations across domains where authority may drift independently.The decomposition is intended to expose a specific failure mode, not to claim exhaustive or perfectly sharp enterprise ontology boundaries.
  • 3.2 The Authority Relation as a cross-domain primitive: The Authority Relation is defined as AR = (Actor, Action, Object, Domain-Context, Justification-Chain, DNA-Coefficient), binding authority across domains.Each element captures who acts, what act is permitted or required, what entity is affected, where authority applies, why it exists, and how practice diverges from specification.
  • 3.2 The Authority Relation as a cross-domain primitive: Delegated human and non-human agents hold authority through distinct AR instances, while Domain-Context scopes authority to a jurisdiction, business unit, system boundary, or similar bounded context.An AR without explicit Domain-Context is underspecified, and delegated authority is treated as a first-class relation rather than collapsed into the delegator’s authority.
  • 3.2 The Authority Relation as a cross-domain primitive: The DNA-Coefficient measures divergence between an AR as formally specified and how it is exercised in practice, and is identified as the framework’s genuinely novel element.The measure is calibrated and empirically estimated within a specific organization.
  • 3.3 Worked illustration: The worked discount-approval case shows divergence across social, business, process, machine, and real-world domains despite a formally specified threshold-based approval relation.Examples include informal escalation to a former incumbent, inconsistent discount calculations, retroactive workflow logging, stale permissions, and delayed margin visibility.

4. The DNA-Coefficient – Toward a measurement methodology

This section proposes the DNA-Coefficient as a research-and-practice methodology rather than a settled instrument, combining three independent measures of divergence between documented and lived authority. It extends measurement to agent-design authority and requires business, technical, and governance affirmation before scores are treated as valid for consequential use.

  • 4.1 Three converging instruments: The methodology combines three independent instruments because no single method reliably captures both documented and lived authority.They are decision-log divergence analysis, organizational network analysis, and structured elicitation.
  • 4.1 Three converging instruments: Decision-log analysis compares documented approvers with actual approvers or de facto decision-drivers reconstructed from records, correspondence, or workflow audit trails.The proposed output is a divergence rate with a directional indicator showing whether actual authority sits differently from documented authority.
  • 4.1 Three converging instruments: Organizational network analysis estimates where influence concentrates by comparing communication and collaboration networks with formal organizational structure, while structured elicitation surfaces tacit trust- and reputation-based influence.Network analysis raises substantial data-governance, privacy, and works-council consultation constraints, particularly in jurisdictions with strong codetermination norms.
  • 4.2 Composite Scoring: The DNA-Coefficient is proposed as a normalized composite divergence value, with zero indicating exact correspondence between documented and lived authority and a directional flag for diagnosis.The framework is intended to flag high-divergence, high-stakes Authority Relations for governance attention.
  • 4.3 The Agent Design Authority Instrument: Agent-design authority requires a fourth instrument tracking who controls an agent’s objectives, action-space boundaries, and runtime Justification-Chain data.The paper applies the same Authority Relation primitive because agent-design authority may exhibit undocumented concentration and remain distributed across engineering, procurement, and vendor-configuration decisions.
  • 4.4 Validation Status and Limitations: None of the four instruments has been empirically tested in this work, and the proposed next step is a single-organization, single-AR-instance pilot.The pilot would specifically use decision-log divergence and agent-design-authority instruments because they carry lower data-governance burden than full network analysis or broad elicitation.
  • 4.5 A Structural Mechanism for Tri-Functional Co-Ownership: A DNA-Coefficient score is not considered valid for agent-permission configuration or governance and audit findings until business, technical, and governance roles independently affirm it.The three checks validate current policy, evidence completeness and integrity, and proportionate use relative to the Authority Relation’s stakes.

5. Agentic AI, the relocation of power and governance implications

Agentic AI relocates the gap between formal authority and actual imperative control to agent design and configuration authority. The framework therefore treats agent authority as an auditable governance, risk-control, and jointly owned enterprise concern.

  • 5.1 A Structural Analogy, Carefully Bounded: Weber’s distinction separates the formal right to issue commands from the probability that actors can secure compliance.The paper uses this structural analogy narrowly, without extending it to Weber’s broader theory of bureaucracy or rationalization.
  • 5.1 A Structural Analogy, Carefully Bounded: Agentic AI relocates, rather than removes, the gap between formal authority and actual imperative control to agent design and configuration authority.The paper contrasts human organizational tendencies with agents that optimize configured objectives and lack an analogous tendency.
  • 5.2 Governance Implications: Agent-design authority should be tracked and audited as a high-stakes Authority Relation, including when configuration depends on vendor defaults and individual engineering decisions.Documenting human approval hierarchies does not reduce governance exposure if agent configuration remains informal and undocumented.
  • 5.2 Governance Implications: Governance design should separate the function controlling agent-design authority from the function benefiting most when that authority’s exercise remains unscrutinised.The paper identifies senior technology executives as natural candidates for significant authority while rejecting their unchecked concentration as a governance design.
  • 5.2 Governance Implications: As agentic AI scales, regulators and auditors are likely to ask who actually had decision authority and whether it was constituted, documented, and current.An auditable Authority Relation with a documented DNA-Coefficient is presented as strengthening defensive and proactive deployment capabilities.
  • 5.4 Why this cannot be the Chief Information Officer’s problem alone: Enterprise adoption should make business, technology, and risk/compliance leadership jointly accountable through a governance structure reporting to business and risk or audit leadership.This proposal extends the co-ownership mechanism and rejects sole ownership of the overall agentic AI investment program by the CIO or equivalent technology leader.

6. Conclusion

The conclusion presents ARF as a cross-domain authority relation spanning Social, Business, Process, Machine, and Real-world domains, while emphasizing organization-specific variation. It frames the DNA-Coefficient and the framework itself as propositions requiring empirical validation, comparison, testing, and critique.

  • Framework contribution: ARF treats authority and decision rights as relations resolved consistently across five domains rather than as properties within one enterprise-ontology domain.The domains are Social, Business, Process, Machine, and Real-world; lived authority varies by organization beyond what a generic template captures.
  • Framework contribution: The DNA-Coefficient is proposed, not yet validated, to measure organization-specific divergence and authority implicit in agent design.The conclusion argues that agentic AI deployment relocates rather than removes the gap between formal authority and actual imperative control.
  • Structural design: ARF is designed without a single loadbearing domain, instrument, or function, using cross-domain interaction to expose divergences and constrain unilateral authority claims.Decision-log gaps may be exposed through organizational network analysis or structured elicitation, while authority claims require business, technical, and governance affirmation.
  • Future work: The framework is offered as a research and practice agenda whose key open items are a validated DNA-Coefficient pilot and a verified full-text comparison with an organizational policy ontology.The proposed comparison concerns Weigand, Johannesson, and Guizzardi’s ontology and the boundary between what ARF depends on and what it adds.
  • Evaluation and disclosure: The conclusion invites researchers, practitioners, and enterprise vendors to test, critique, extend, or refute ARF through cumulative, citation-based scholarly and governance norms.The author also discloses language-model assistance for literature search, drafting suggestions, editing, readability, and manuscript organization under the author’s direction.
Loading 2608.15832v1…