Moral Rights and Generative AI Training under the Pakistani Copyright Ordinance 1962: A Comparative Analysis

Abstract

The normative foundations of copyright regard moral rights as an extension of the author's personality rather than a tradable commodity. This envisages conferring upon authors a range of rights that continue to apply even after their economic rights are transferred or exhausted. The training of generative AI on extensive datasets of copyrighted works has attracted significant scholarly interest regarding potential violations of economic rights. However, less attention has been devoted to assessing how moral rights are implicated by large-scale data acquisition and machine learning processes used in generative AI training under the laws of developing states such as Pakistan. This research aims to fill this gap by employing a qualitative doctrinal methodology to analyse moral rights-related concerns arising from AI training under Pakistan's Copyright Ordinance 1962. It seeks to identify unresolved concerns regarding moral rights during data collection, preprocessing, and fine-tuning stages of the training process. The study conducts a comparative analysis of the Copyright Ordinance with moral rights frameworks of the United Kingdom and the United States. Finally, the normative position of moral rights issues within the laws of these jurisdictions is assessed using the personality theory as a normative framework. The study finds that attribution-based claims provide a stronger legal basis for challenging AI training activities than integrity-based claims due to the substantial evidentiary burden in proving prejudice to honour or reputation. The study further demonstrates that Pakistan’s moral rights framework offers broader protection than the United States but lacks specific guidance regarding its application to AI training.