Training Data as Extraction AI, Traditional Knowledge, and the Sovereignty Gap in Global IP Law Ayush Tripathi

Training Data as Extraction: AI, Traditional Knowledge, and the Sovereignty Gap in Global IP Law

The rapid proliferation of large-scale artificial intelligence systems has generated a structural tension at the heart of global intellectual property law. AI models are trained on vast datasets that routinely incorporate traditional knowledge, indigenous cultural expressions, and community-generated content from the Global South without consent, attribution, or benefit-sharing. Yet existing IP frameworks, designed around individual authorship and formal documentation, provide no cognizable protection against this form of digital extraction.

This paper argues that AI training practices constitute a novel mode of knowledge appropriation that existing international IP instruments are institutionally incapable of addressing. The TRIPS Agreement's minimum standards regime, the Berne Convention's originality threshold, and even the CBD's Nagoya Protocol on access and benefit-sharing collectively fail to create enforceable entitlements for communities whose knowledge is absorbed into commercial AI systems. The resulting asymmetry is structurally colonial: Global South knowledge enriches Global North AI infrastructure while generating no reciprocal obligation.

The paper proposes reconceptualising training data governance through the lens of digital sovereignty, arguing that states must assert regulatory jurisdiction over the extraction and commercial use of culturally embedded data originating within their territories. Drawing on emerging multilateral discussions at WIPO's Intergovernmental Committee on Genetic Resources, Traditional Knowledge and Folklore, it outlines a minimum framework for mandatory disclosure, prior informed consent, and equitable benefit-sharing applicable to AI training pipelines. Without such reform, global equity in the digital age remains aspirational rather than operational.

Keywords: Traditional knowledge, AI training data, digital sovereignty, TRIPS, benefit-sharing