Same Data, Different Rules: Copyright, Data Protection, And The Training Of Generative AI Models In The United States, The European Union, And India
Akshat Rawal, Manav Rachna University
ABSTRACT
Generative AI systems are commonly trained on very large collections of online material. Those collections may contain both copyrighted works and personal data, bringing the same training activity under two areas of law. Copyright law governs the use of protected works; data-protection law governs the processing of information about people. Each has its own rules on permission, exceptions, and remedies.
This article compares the United States, the European Union, and India. It argues that copyright and data-protection requirements operate cumulatively: meeting one does not establish compliance with the other. The two fields intersect most clearly when developers assess where training data came from and whether a model can retain or reveal information from that data. The article also examines licensing proposals, particularly India’s proposed mandatory blanket licence. It argues that licensing may address disputes over copyrighted works, but cannot by itself resolve privacy obligations owed to individuals. The analysis distinguishes established law from interpretation and flags authorities that require further verification.
Keywords: generative AI; text and data mining; fair use; GDPR; Digital Personal Data Protection Act; training data; memorization; licensing; machine-readable opt-out.
