AI Data Scraping and Fair Dealing Exception Under the Copyright Act

AI companies train their systems on huge amounts of data collected from different writing, images and music copied from the internet, often without asking or paying the people who made it. This is not only a copyright problem in the present it is also a trade problem. Because when the data crosses the borders it has a market price and is treated differently in different countries. India has no law that specifically allows text and data mining so courts fit AI training sthe Delhi High Court did this for the first time, at the interim stage and it was held that training of large language models on openly published news content was fair dealing. This article asks three questions is that reading of the court correct, does it meet the three step test mentioned in Article 13 of the Agreement on Trade Related Aspects of Intellectual Property Rights and what are the reforms India need at the present scenario. It uses doctrinal and comparative methods, drawing on Indian case law, World Trade Organization panel reports, decisions from the United States and United Kingdom cases and rules in the European Union. Most of the existing discussion treats the issue only as a question of Indian fair dealing. This article looks at it through India's trade obligations as well, and asks what creators should receive when their work is used to build and train commercial AI products. It argues that creators deserve payment, and that Article 13of TRIPS points the same way. It suggest an amendment to the Copyright Act that permits AI training on lawfully accessed works, lets owners opt out, requires commercial developers to pay royalties through a collective system.