A Control-Based Differentiated Framework for the Content Governance Obligations of Generative Artificial Intelligence Service Providers

Vollständiger Abstract

Worum geht es in dieser Arbeit?

The content governance obligations of generative artificial intelligence (GAI) service providers constitute a central issue in contemporary AI regulation. Existing legal frameworks, however, impose substantially uniform content governance obligations on GAI service providers without adequately accounting for the significant differences among various service models. Drawing upon risk-control theory, this Article argues that the content governance obligations of GAI service providers should be differentiated according to the degree of substantive control they exercise over the content-generation process. Based on differences in such control, GAI service models may be categorized into three types: pure prompt-based generation services, retrieval-augmented generation (RAG) services, and foundation model fine-tuning services. Correspondingly, each category should be subject to a distinct set of content governance obligations calibrated to its respective level and form of control. A differentiated regulatory framework that allocates content governance obligations on the basis of substantive control can better reconcile the dual objectives of safeguarding the security of AI-generated content and promoting technological innovation, thereby achieving a more proportionate and refined balance between regulation and innovation.