THE CONTEXT: The Statute of Anne (1710) birthed modern copyright as a response to the printing press. Each technological leap since — gramophones, photocopiers, MP3, the Internet — triggered claims that the law would “break”. Generative artificial intelligence (Gen-AI) is the latest stop on this 20-year cycle of copyright angst.
THE BACKGROUND:
-
- Unlike past disputes over copies, today the friction point is training Large Language Models (LLMs) that ingest billions of tokens, including copyrighted works, to learn statistical relationships, rather than replicating the works verbatim. Indian publishers ANI Media and the Federation of Indian Publishers have sued OpenAI before the Delhi High Court precisely on this ground — forcing the court to ask whether storing data for training by itself is infringement and whether Section 52 “fair dealing” can shelter it.
- Unlike traditional copyright infringement, which involves reproducing copyrighted content, AI training involves processing copyrighted content to generate new outputs, raising questions about consent, ownership, and fair use. Globally, lawsuits against AI companies underscore the urgency of reforming copyright laws. In India, the Copyright Act, 1957, governs intellectual property, but its provisions, rooted in pre-digital paradigms, struggle to address AI’s complexities.
INDIA’S COPYRIGHT LANDSCAPE:
India’s Copyright Act, 1957, amended in 2012, includes provisions for digital technologies but lacks clarity on AI training. Section 52 outlines fair-dealing exceptions, which are narrower than the U.S. fair use, covering specific acts such as private use or research, but not broad data mining. Recent developments include:
-
- Litigation Surge: The Federation of Indian Publishers and Asian News International filed cases against Open AI in 2025, alleging unauthorized use of their content for AI training.
- Judicial Observations: Amicus curiae Professor Scaria emphasized the feasibility of “unlearning” copyrighted content and the need to ensure access to legitimate information.
- Industry Response: Open AI’s opt-out mechanism allows creators to exclude their works from future training but does not address past data use, raising fairness concerns.
GLOBAL PERSPECTIVE: COMPARATIVE POLICY FRAMEWORKS
-
- United States: The U.S. Copyright Act’s fair use doctrine allows for flexible interpretation, but recent U.S. Copyright Office reports suggest that commercial AI training may not qualify as fair use, siding with content creators.
- European Union: The EU’s Copyright Directive (2019) permits Text and Data Mining (TDM) for research but requires rightsholders’ consent for commercial purposes, with an opt-out option.
- Singapore: Recent guidelines (2024) clarify that AI training on copyrighted material is permissible if it aligns with fair use principles, emphasizing transparency.
- Japan: Japan’s Copyright Act allows TDM for non-commercial purposes, fostering AI innovation while protecting creators through licensing agreements.
POLICY FRAMEWORK IN INDIA: EXISTING PROVISIONS
-
- Copyright Act, 1957: Governs intellectual property with provisions for fair dealing (Section 52) and digital rights management (Section 65A) but lacks AI-specific clauses.
- Information Technology Act, 2000: Addresses digital content but does not cover AI training or data scraping.
- National AI Strategy (NITI Aayog, 2018): Promotes AI innovation but lacks guidelines on intellectual property, focusing primarily on sectors such as healthcare and agriculture.
ECONOMIC STAKES
-
- Global AI market projected at USD 320-380 bn by 2027; Gen-AI will command 33 %.
- Indian publishing still matters; the book market revenue is USD 10.4 bn (₹ 86,000 cr) in 2024, forecast to hit USD 14.6 bn by 2030.
- Nasscom-BCG (2024) estimates Gen-AI could add USD 60-65 bn to India’s GDP by 2030 through productivity gains, vernacular interfaces, and new service exports.
THE ISSUES:
-
- Legal Ambiguity: The Copyright Act, 1957, does not address AI training, and fair dealing exceptions are too restrictive, limiting innovation and creator protections.
- Data Memorization: AI models may inadvertently reproduce copyrighted content, violating creators’ rights, as seen in global cases like New York Times vs. Open AI.
- Jurisdictional Challenges: Open AI’s argument that Indian courts lack competence complicates enforcement, highlighting gaps in international copyright harmonization.
- Economic Disparities: Large AI firms with substantial resources can negotiate licensing deals, while smaller Indian startups struggle, resulting in an uneven playing field.
- Access vs. Protection: Overly restrictive copyright laws could limit AI’s access to quality data, stifling innovation, while lax regulations may harm creators’ livelihoods.
- Ethical Concerns: Unauthorized use of copyrighted content raises moral questions about attribution and fairness, particularly for India’s creative industries.
THE CHALLENGES:
-
- Limited AI Literacy Among Judges and Policymakers: A 2024 survey by the Vidhi Centre for Legal Policy found that only 12% of Indian judges reported familiarity with AI technologies, compared to 45% in the EU.
- Conflicting Interests of Publishers, Startups, and Consumers: Publishers argue that AI training on their copyrighted works without consent or compensation undermines their economic and moral rights, as protected under Section 14 of the Copyright Act, 1957. Restrictive copyright laws could stifle innovation, as highlighted by Nasscom’s 2025 report, which notes that 70% of Indian AI startups lack funds for data licensing.
- Experimental Model Unlearning and Resource Constraints for SMEs: Unlearning copyrighted content from AI models is a technically experimental and resource-intensive process, particularly for small and medium-sized enterprises (SMEs), which hinders the equitable implementation of policies. The U.S. Copyright Office’s 2025 report flags unlearning as “promising but unproven,” while EU firms, such as Hugging Face, are experimenting with it, highlighting India’s lag in R&D.
THE WAY FORWARD:
-
- Legislative Reforms: Amend Section 52 of the Copyright Act, 1957, to include a “Computational Use” exception for Text and Data Mining (TDM), allowing right-holders to opt out via a “no-AI” flag, and introduce a statutory Extended Collective Licensing (ECL) system with a royalty clearing-house. This ensures fair remuneration for creators while enabling AI innovation. The reform aligns with the Berne Convention’s three-step test, striking a balance between access and protection.
- Technological Interventions: Develop a government-funded “Unlearning Toolkit” for small and medium enterprises (SMEs) to remove copyrighted content from AI models and mandate a dataset provenance and transparency registry for frontier models with over one billion parameters. This levels the playing field and enforces accountability. The toolkit reduces compliance costs, fostering technological equity.
- Institutional Mechanisms: Establish an AI-Copyright Task Force under the Ministry of Electronics and Information Technology (MeitY) and the Department for Promotion of Industry and Internal Trade (DPIIT) and create specialized Intellectual Property-Technology (IP-Tech) Benches in High Courts. These ensure speedy dispute resolution and coherent policy formulation. The task force fosters whole-of-government
- Capacity Building: Launch a joint certificate program on “AI and Intellectual Property” by the National Academy of Legal Studies and Research (NALSAR) and the Lal Bahadur Shastri National Academy of Administration (LBSNAA), alongside online Massive Open Online Courses (MOOCs) for creators on opt-out tools. This enhances human resource readiness. It supports digital skilling and civil-service reform.
- International Collaboration: Lead a WIPO proposal for a “Fair Learning” Declaration recognizing developmental needs in AI training and negotiate mutual-recognition clauses with the EU and U.S. for dataset licenses. This shapes global norms and reduces forum-shopping. It enhances tech diplomacy and digital multilateralism.
- Startup Support: Provide Compute and Corpus Vouchers through the IndiaAI Mission for SMEs and create a Public Domain Megacorpus of Indian-language texts under Creative Commons licenses. This democratizes innovation and protects linguistic diversity. It aligns with Atmanirbhar LLMs and public digital infrastructure.
- Ethical Frameworks: Embed copyright attribution, bias testing, and watermarking in a Responsible-AI “Blue Tick” Certification linked to government procurement preferences. This encourages market-led compliance and ethical AI development. It promotes soft-law nudges and ethical by design
THE CONCLUSION:
A “fair learning with fair returns” framework that legalises computational analysis, guarantees creator remuneration, mandates dataset transparency, and equips courts with techno-legal capacity will convert today’s courtroom skirmishes into tomorrow’s innovation dividend. In doing so, India can model a digital constitutionalism—protecting the right to create, the right to innovate, and the right of society to learn.
UPSC PAST YEAR QUESTION:
Q. The application of Artificial Intelligence as a dependable source of input for administrative decision-making is a double-edged sword. Discuss. 2024
MAINS PRACTICE QUESTION:
Q. Enumerate the key issues arising from AI’s use of copyrighted material for training and suggest a comprehensive policy framework to address these issues, ensuring inclusive growth and global competitiveness.
SOURCE:
https://www.thehindu.com/opinion/op-ed/copyrights-tryst-with-generative-ai/article69590505.ece
Spread the Word