When Compliance Becomes Exclusion: Proportionality, Liability Legitimacy, and the Governance of Third-Party AI Fine-Tuning

Abstract

Third-party fine-tuning is consequential in AI governance, yet current compliance obligations—designed for large firms—impose disproportionate burdens on SMEs, creating structural exclusion rather than managing risk. This article argues that proportionality must serve as the normative foundation for liability legitimacy. By analyzing EU risk-tiered frameworks, the fragmented American approach, and China’s "regulate through support" model, the study identifies a need for differentiated obligation structures. The article proposes a liability architecture comprising four elements: duty structures calibrated to deployment risk; application-side regulatory thresholds; graduated transition mechanisms for institutional growth; and safe harbor protections conditioned on participation in collective governance structures. Crucially, it advocates for the construction of public compliance infrastructure—shared audit pools, standardized templates, and centralized ethics reviews—to transform compliance from an individualized, fixed-cost burden into a shared institutional service. Only by grounding liability in proportional governance can the AI ecosystem remain epistemically and socially robust.

Keywords