Fair Use of Copyrighted Material by Generative AI Models

Recent U.S. court rulings involving Anthropic and Meta recognize the transformative use of copyrighted material by generative AI models as fair use. This evolving legal stance is grounded in the purpose of training large language models (LLMs). While India does not have a codified fair use doctrine akin to the U.S., there are corresponding legal provisions and ongoing debates relevant to Indian copyright law in the context of AI.

FAQs

1. What is fair use and how is it being applied to generative AI in the U.S.?

Fair use in the U.S. allows limited use of copyrighted material without permission for purposes such as commentary, teaching, or research. Courts have recently extended this to include AI training, particularly when the use is highly transformative, such as converting text into machine-readable tokens without reproducing content for human consumption.

2. What factors do U.S. courts consider in AI-related fair use cases?

Courts assess:
• The purpose and character of the use (commercial vs. transformative)
• The nature of the copyrighted work
• The amount and substantiality of the portion used
• The impact on the market for the original work
In the Meta and Anthropic cases, courts recognized that although entire works were ingested, the use was non-expressive and transformative for LLM training rather than replicating original content.

3. What are the key legal takeaways from the Meta and Anthropic rulings?

• Training LLMs on copyrighted text can qualify as fair use if the output does not reproduce or compete with the original work.
• Transformation of content into machine-readable tokens is considered non-substitutive.
• Licensing arguments are secondary when weighed against fair use principles.

4. Does Indian copyright law have a fair use clause?

India follows the concept of fair dealing under Section 52 of the Copyright Act, 1957. Fair dealing is narrower than U.S. fair use and applies only to specific purposes such as private research, criticism, or education. Indian law does not include a flexible test for transformative uses like AI training.

5. How would India handle the training of AI on copyrighted data under current law?

• Indian law does not explicitly permit AI training on copyrighted material without the author’s consent.
• Section 52(1)(a) may protect usage for research and educational purposes, but commercial AI training likely falls outside this safe harbor.
• Legislative reform or judicial clarification may be required to address AI-specific use cases.

6. What are the possible legal challenges in India?

• Authors and publishers could challenge AI developers for unauthorized use of copyrighted books, articles, or digital content.
• Ambiguity around what constitutes transformative use could result in injunctions or lawsuits.
• Legal disputes may raise broader constitutional questions about innovation vs. copyright protection.

7. How can India modernize its copyright framework to accommodate generative AI?

• Introduce a structured fair use test, similar to the U.S., particularly for AI and research purposes.
• Clarify whether machine learning and data ingestion qualify as valid research or educational use.
• Develop a licensing framework or safe harbor mechanism for AI developers using public-domain content or opt-out material.

8. Are there any Indian judgments or policy discussions on this yet?

• Indian courts have not yet ruled on generative AI and copyright issues.
• Initiatives such as the National Digital Library and Digital India reflect growing interest in balancing intellectual property rights with digital access.
• Ongoing parliamentary discussions on AI regulation may eventually provide clarity on this emerging legal area.