Section 52 has narrow fair‑dealing rules that does not clearly cover large‑scale commercial AI training, calling massive dataset ingestion “research” is a stretch and it does not fit the narrow meaning of research in the law and could weaken creator’s motivation to produce new work.
Generative AI models are trained on huge amounts of text, images and other creative works, many of which are copyrighted. People are arguing about whether AI companies need permission to use these works for training, or whether the law already allows it. In India, the law isn’t clear on this. At first sight, people look to Section 52 of the Copyright Act, 1957, which lists when using copyrighted material is allowed without permission, but Section 52 doesn’t clearly cover big commercial AI training, it mainly protects small-scale uses like study, teaching, or research.
In India we don’t have the U.S. style “fair use” where courts can allow almost anything if it seems fair. We have “fair dealing” under Section 52, which is fixed. It only covers things like private use, criticisms, or news reporting. Because the list is specific, courts are expected to stick to the list. They should not add new exceptions just because AI is useful.
In the Oxford v. Rameshwari Photocopy case, the Delhi High Court kept Section 52 narrow for education. It said, we won’t expand the list just because photocopying got easier. That shows Indian courts are careful about not creating brand-new exceptions for new technology.
Some AI companies say training is “research” so it should be allowed. But there’s a difference that is humans doing research usually read parts of an article, not copy the whole thing and AI training often needs full copies or large chunks of text, images, etc. to learn patterns. Also, most LLMs are commercial products they are sold as subscriptions or services to make money. Section 52 is meant for private study or non-commercial research, not large-scale profit-making systems.
If courts allow AI companies to train on copyrighted books, articles, music, etc. without permission or payment, it affects authors, journalists, musicians, publishers, researchers. Their work could be used for free to build AI that competes with them. If Section 52 is read broadly Indian AI startups can train models without buying expensive data. That could help the AI industry grow fast. If it is read broadly for commercial AI Creators may lose control and income. That could discourage people from making new books, music, or journalism.
Right now, no one is sure what is lawful. AI companies don’t know if they will be sued later or not. Creators don’t know if the law will protect their work. That uncertainty will make them accountable for cost and slows down decisions for both technology and creative industries.
Section 52 was written long before there exists any AI. It does not say anything about text or training AI on huge amounts of data. So, courts should not just extend it to cover AI. Instead of forcing the old laws and putting into new rules, lawmakers are in a better position to make a new and transparent law. They can create rules that will balance both AI innovation and creator’s rights properly. Until and unless Parliament makes a new law, or courts give a clear law, AI companies should not rely on the training of copy righted works.
Section 52 was made for people, schools, and small-scale research where it is allowing the large-scale, commercial AI training under it would be a huge change, not a small one. It would affect both tech companies and artists, writers, musicians, etc.
In the end, Section 52 has narrow fair‑dealing rules that does not clearly cover large‑scale commercial AI training, calling massive dataset ingestion “research” is a stretch and it does not fit the narrow meaning of research in the law and could weaken creator’s motivation to produce new work. Rather than relying on broad judicial readings, India needs targeted legislation with clear exceptions, licensing, and remuneration, to balance innovation with fair pay for authors.