The Arbitration Legitimacy Stack: Due Process, Explainability, and Enforceability of AI-Augmented Awards in a Post-AGI World

Haley Yi, David Scott Lewis

Track: Track 2: Socio-Economical and Future Visions

Keywords: Arbitration Legitimacy Stack, AI-Assisted Award Record (AAR), AI-assisted arbitration, award enforceability, due process, rule-level transparency, inference-level transparency, narrative-level transparency, process evidence, provenance metadata, solver-checkable traces, prompt injection, explanation laundering

TL;DR: As arbitration adopts AI, enforceability will hinge on auditable process, not just outcomes. Proposes a 3-layer legitimacy stack (rule/inference/narrative) and an AI-Assisted Award Record with provenance, traces, and override logs.

Abstract: International commercial arbitration is a private, transnational governance system whose legitimacy rests on a fragile bargain: parties trade public adjudication for speed and expertise, but they still demand due process and an enforceable award. As AI assistance becomes routine in case management, evidence triage, translation, and draft-award generation, enforceability will increasingly hinge on process evidence—what a tribunal can show about how it reached a decision, not just what it decided. We propose the Arbitration Legitimacy Stack, a socio-legal framework that decomposes legitimacy into three auditable layers: (1) rule-level transparency (which rules and instruments were applied), (2) inference-level transparency (inspectable steps from premises to conclusions, ideally machine-checkable for rule-governed determinations), and (3) narrative-level transparency (human-readable reasons that faithfully summarize the inference artifacts). We argue that post-AGI arbitration fails when these layers decouple: fluent narratives can mask untraceable inferences, while formal traces can become unintelligible to parties. To operationalize the stack, we introduce an AI-Assisted Award Record (AAR): a compact bundle containing AI-use disclosure, provenance metadata for evidence operations, solver-checkable traces for formalizable steps (e.g., jurisdiction, admissibility, timelines, costs), and an override log capturing non-delegable human judgment. We sketch confidentiality-preserving disclosure modes and outline a Tiny-Paper-sized research agenda: micro-benchmarks tied to procedural rulebooks, cross-layer faithfulness metrics, and auditability-by-design patterns resilient to prompt-injection and explanation-laundering attacks.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Submission Number: 27