Intellectual Property and Personal Data in AI Datasets Under India's DPDP Act 2023

Abstract

The rapid expansion of generative AI has challenged India’s fragmented legal regime governing AI training data, spanning the Copyright Act, 1957, trade secret protection, and the Digital Personal Data Protection Act, 2023 (DPDPA). This study explores the doctrinal incompatibility between India’s purpose-specific fair dealing framework under Section 52 and the industrial-scale reproduction intrinsic to AI training, which fails the jurisdictional “purpose test” articulated in Super Cassettes Industries Ltd v. Hamar Television Network (2011). The study exposes the structural inadequacy of trade secret law in protecting the “composited value” of large-scale aggregated datasets, which lack the identifiability and durability required for conventional protection. The DPDPA’s consent-centric architecture is functionally unworkable in billion-token training corpora characterized by attenuated data-principal relationships. Concrete doctrinal fault lines, including uncertainty surrounding “reproduction in material form” under Section 14(a)(i), transparency-trade secret conflicts identified in the DPIIT Working Paper on Generative AI and Copyright, and cross-border transfer constraints under Section 17 of the DPDPA have been mapped. Legal uncertainty will undermine both AI innovation and stakeholder protection in India if an integrated statutory framework for permissible training practices and rights allocation is not opted.