While generative artificial intelligence has made production of text, images, music or other software code and audiovisual works more efficient than ever before, it has also disrupted the foundational structure of copyright law. Such concerns lead to three interrelated questions in Indian copyright doctrine, as follows: does the use of protected works for training purposes infringe rights, when (if ever) do AI-generated outputs constitute infringing copies or adaptations, and can the mere imitation of a human creator’s style be legally determined without enclosing ideas, methods and aesthetics? Despite containing provisions on computer-generated works, exclusive rights and infringement, fair dealing and moral rights, The Copyright Act 1957[1] was not drafted to address large scale model training or probabilistic content generation. This paper is doctrinal in nature, inspecting Indian statutes, constitutional provisions as well as case law and policy pertaining to international obligations under copyright laws and comparative jurisdictions including the European Union, Japan and the US. It contends that Indian law ought to reject both extremes: otherwise-free licence for training on protected data, and an over-broad copyright veto over all machine learning. Instead, India should at least move towards a careful calibrated framework that divides copying that takes place during training, substitutive output stage as well as author-specific creative imitation. This paper suggests a new exception for text-and-data-mining by establishing lawful access and transparency/opt-out safeguards through compulsory or collective licensing, plus a human creative control test on AI-assisted authorship, followed by a fault-sensitive liability model for developers, deployers and users.
[1]Copyright Act 1957 (India) ss 2(d)(vi), 13, 14, 17, 51–55B.
Publication Date: 2026-06-13