UPSC Relevance
- GS Paper 2: Governance, Legal frameworks, and Rights issues
- GS Paper 3: Science and Technology – Developments and their applications
Essay Topics: Ethics of Artificial Intelligence, Copyright in the Digital Age
Why in News?
Two recent summary judgments in the United States—Andrea Bartz vs. Anthropic PBC and Richard Kadrey vs. Meta Platforms Inc.—have brought global attention to the application of the fair use doctrine in the context of Generative AI (GenAI) training. These judgments could shape the future of copyright law in the AI era.
Background: Copyright and AI Training
To develop Large Language Models (LLMs) like GPT, Claude, and LLaMA, companies rely on enormous datasets. These datasets often contain:
- Public domain works (no copyright)
- Copyrighted materials, including books and articles
- Illegally sourced content, like pirated e-books
This raises a crucial legal question:
Can copyrighted works be used without permission for AI training?
In the U.S., the answer depends on the “fair use” doctrine, a legal exception that allows limited use of copyrighted content under specific conditions.
What is Fair Use? – The Four-Factor Test
In U.S. law, fair use lets people use copyrighted material without permission in certain situations — like teaching, research, or commentary. To decide if something counts as fair use, courts look at four key factors:
1. Purpose and Character of Use
- Is the use transformative — does it change the original or use it for a new purpose (e.g., training an AI)?
- Non-commercial or educational uses are more likely to qualify.
2. Nature of the Original Work
- Factual content (like textbooks or news) is more likely to be fair use than creative works (like novels or movies).
3. Amount and Substantiality
- How much of the work was used?
- Was the heart or most important part taken?
4. Effect on the Market
- Does the use reduce the market value or sales of the original?
- Could it replace the original in the market?
Courts look at these case by case, but transformative use and market harm are usually the most important.
Case 1: Anthropic (Claude AI)-What Happened?
Anthropic used books (some bought legally, others pirated) to train its AI model Claude.
Court Decision:
- Fair use allowed for books that were legally purchased and digitised.
- Fair use denied for pirated books.
Judge’s View:
- Training AI is not a threat to creative competition.
- It’s like teaching students using books — not stealing creativity.
Case 2: Meta (LLaMA AI)
What Happened? Meta allegedly downloaded pirated books to train its AI model LLaMA.
Court Decision:
- Training AI was highly transformative, so it could qualify as fair use.
- Plaintiffs failed to show real market harm.
Judge’s Warning:
- Even if transformative, AI could harm authors indirectly, by flooding the market with AI-generated content.
- The case will continue on whether Meta illegally distributed pirated material during training.
| What is Fair Use (Fair Dealing)?(in INDIA) Fair dealing, as defined under Section 52(1) of the Copyright Act, 1957, allows limited use of copyrighted material without permission. It is applicable when the use is for educational, critical, journalistic, or research purposes, balancing the rights of the creator with public interest. Criteria for Fair Use (Qualitative Factors): Purpose of Use: If the material is used to inform, educate, or critique—like in journalism or parody—it’s more likely to be fair. However, using content for commercial gain or misleading purposes is not considered fair dealing. Nature of Work: It’s easier to use factual, published, or publicly accessible works than unpublished or highly creative ones. Amount Used: Using a small, necessary portion of the work generally increases the chances of fair use. But even short clips could still infringe if they capture the essence of the original work. Market Impact: If the use of the material harms the original’s ability to earn money, replaces the original, or attracts its audience, it’s not considered fair dealing. The greater the financial harm to the copyright holder, the less likely the use qualifies as fair. Example: In the case of TV Today vs NewsLaundry, limited use of video clips was allowed under fair use because it did not cause financial harm or reduce the value of the original broadcast. |
Comparative Analysis of Judgments: Anthropic vs Meta
| Aspect | Anthropic Case | Meta Case |
| Transformative Use | Recognised as transformative | Recognised as transformative |
| Use of Illegal Sources | Fair use denied for pirated books | Not directly assessed |
| Market Harm | Dismissed by court | Plaintiffs failed to prove harm |
| Additional Proceedings | Yes – case continues on illegal sources | Yes – case continues on distribution issue |
Key Takeaways:
- Both courts agree that training AI is transformative, meaning it gives the original work a new purpose (like teaching a machine).
- Anthropic was penalised for using pirated books, showing a stricter stance on illegal sourcing.
- In Meta’s case, the issue of pirated content was not fully examined yet, but the court flagged possible concerns about market dilution.
- Neither court found strong evidence that AI training caused direct market loss — but future AI-generated content flooding the market is a concern.
Case That Rejected Fair Use: Thomson Reuters v. Ross Intelligence
What Was the Case About?
● Ross Intelligence used legal content (from Westlaw, owned by Thomson Reuters) to train an AI tool that retrieved and served court judgments.
● Unlike generative AI, this system didn’t create new content—it just repackaged existing legal texts.
Court Ruling:Not Fair Use
Why Was Fair Use Rejected?
- No Transformative Use:
○ The AI didn’t add new meaning or function — it just republished legal data.
○ It competed directly with Thomson Reuters’ product. - Market Harm:The AI tool substituted the original product, hurting its commercial value.
Key Insight:When AI is used to simply copy and compete—not to transform—courts are less likely to allow fair use.
Broader Implications of Fair Use in Generative AI
1.Favourable Trend for GenAI Firms
● Courts are leaning toward fair use if:
○ The AI use is highly transformative, and
○ It doesn’t harm the original market directly.
2.Challenges for Authors and Copyright Owners
● It’s hard to prove market harm, especially when:
○ GenAI outputs are not identical to the original work.
○ The damage is indirect, like market dilution.
3.Importance of Data Sourcing
● Using pirated or illegally obtained content can:
○ Invalidate fair use claims, as seen in the Anthropic case.
○ Invite legal consequences even if the AI use is transformative.
4.Copyright Law Still Evolving
● Current rulings are preliminary and vary by jurisdiction.
● New cases will shape how AI and copyright law interact in the future.
Way Forward
1.Need for Clear Guidelines
● Governments and courts must issue transparent rules on:
○ What kind of data can be used for AI training?
○ When and how fair use applies.
2.Fair Compensation for Creators
● Licensing models can ensure:
○ Creators and publishers get paid when their work trains AI.
○ A sustainable ecosystem for both AI development and content creation.
3.🇮🇳 Indian Context
● India must explore whether Section 52 of the Indian Copyright Act:
○ Is strong enough to handle AI training-related disputes.
○ Needs amendments or clearer judicial interpretation.
4.Global Coordination Needed
● The world needs a shared framework to balance:
○ Innovation in AI
○ Protection of intellectual property
A global dialogue on AI and copyright is essential to avoid exploitation while promoting technological growth.
Conclusion
The application of the fair use doctrine to Generative AI marks a crucial intersection of law, technology, and creativity. While recent U.S. court rulings in favor of Meta and Anthropic signal judicial recognition of the transformative potential of AI training, they also underscore unresolved issues such as illegal sourcing and market impact. As AI continues to reshape the content economy, striking a balance between innovation and intellectual property rights will be essential. Moving forward, courts, policymakers, and creators alike must engage in a nuanced dialogue to build a fair, transparent, and future-ready legal framework for the age of artificial intelligence.
| How is Global AI currently governed? India: NITI Aayog has introduced guiding documents like the National Strategy for Artificial Intelligence and the Responsible AI for All report, focusing on social inclusion, innovation, and trustworthiness. United Kingdom: Adopts a light-touch approach, urging sector-specific regulators to apply existing regulations and published principles for safety, fairness, transparency, and accountability. US: Released a Blueprint for an AI Bill of Rights, emphasizing sector-specific governance and principles to address economic and civil rights harms. China: In 2022, China introduced the world’s first binding regulations on AI, focusing on recommendation algorithms and their impact on information dissemination. |
✍️ Mains Practice Question
Q. In light of recent U.S. court rulings, critically examine the application of the fair use doctrine in training generative AI systems. Discuss its implications for copyright law and creative industries. (250 words)
SOURCE- THE HINDU
Found this helpful?
Bookmark for revision, Practice the mains question, and
Share with fellow aspirants! THANK YOU