Infrastructure-Mediated Multilateralism

A Technical Architecture for International AI Governance

Abstract

International AI governance confronts a paradox: normative consensus on principles like transparency, fairness, and accountability has never been stronger, yet this consensus has failed to prevent ethics washing, capability concentration, or harmful deployments. This paper argues that the problem lies not in insufficient agreement but in absent infrastructure for translating principles into verifiable practice. We propose the Multilateral AI Commons for Peace and Sustainability (MACPS), a five-layer technical architecture addressing three governance deficits: coordination failures across fragmented regulatory regimes; capability asymmetries between Global North and South; and credibility gaps enabling performative compliance. Drawing on Huayan Buddhist philosophy's concept of 'perfect interfusion'—mutual constitution without loss of distinctiveness—we theorise infrastructure-mediated multilateralism: cooperation emerging from shared technical systems that generate material interdependence rather than relying on normative alignment alone. Engaging critically with scholarship on the politics of infrastructure, we argue that technical systems' inevitable embedding of values presents an opportunity for intentional design that encodes equity commitments as architectural constraints. The architecture comprises semantic ontology for standards interoperability, federated compute resources with equity-weighted allocation, privacy-preserving data trusts, mechanism-level evaluation benchmarks, and machine-readable implementation guides. This design paper demonstrates that shared infrastructure can achieve governance objectives that normative frameworks alone cannot, offering a path forward for multilateral cooperation amid geopolitical fragmentation.