This article proposes a two-track framework for using artificial intelligence in international arbitration while preserving enforceability under the New York Convention (NY Convention) and due-process minimums. Track one, Artificial Inteligence (AI)-assisted arbitration, keeps human arbitrators fully responsible for fact-finding, legal reasoning, and the signed award, while using AI for document handling, translation, retrieval, and drafting under disclosure, symmetric access, and strict version control. Track two, AI-exclusive arbitration, treats a certified AI pipeline as the merits decision-maker and is recommended only for narrowly scoped sandbox pilots (highly structured, low-value, high-volume disputes) with explicit consent, frozen configurations, integrity logging, and a human legality/dueprocess backstop. To help courts apply existing refusal grounds without reopening the merits, we introduce a proportional AI Usage and Provenance Dossier (tool/version disclosure, hash manifests, sealed logs, exception reporting, and explicit explainability limits). We connect these operational controls to EU compliance anchors (General Data Protection Regulation (GDPR) and the EU AI Act) and to practical threats (prompt injection, retrieval poisoning, drift, and log omission), emphasizing that cryptographic artifacts provide tamper-evidence for recorded steps, not guarantees of correctness or completeness.
Sienicki et al. (Wed,) studied this question.