Introduction
Technology has long had a disruptive influence on intellectual property law. During the centuries-long history of copyright, its scope has been challenged time and time again by technological breakthroughs-from the use of the printing press, to the advent of photography, to photocopiers and digital peer-to-peer file sharing and web scraping. Laws have never remained the same in the face of these disruptive technologies.
However, this rise of generative AI is a true paradigm shift. Last time, we increased the scale and speed of replication but relied on humans to supply the creativity. Generative AI has turned this prediction on its head. By consuming billions of copyrighted works from the open web, our machine learning systems are now capable of creating new text, images, sounds and code in response to a straightforward natural language prompt.
The move creates an existential dilemma for statutory copyright. When algorithms assemble human information and cultural expression to generate a competitive commercial offering, lawyers, regulators, and judges around the world face the same fundamental issue: Will conventional copyright survive generative AI, or are the premises of copyright law inherently flawed?
The Core Tensions: Ingestion Versus Creation
The conflict between copyright and generative AI takes place on two axes of operation: the Input Frontier (feeding the AI system training data) and the Output Frontier (ownership of the work produced by the system).
1. The Input Frontier: Ingestion, Scraping, and the Limits of Fair Use
In order to build the massive language models (LLMs) or diffusion based image generators, tech companies will mine petabytes of data, from copyrighted books and journalism to digital art, music, and source code. Right’s holders contend it is literally wholesale copying for economic benefit (under copyright’s reproduction rights), while AI developers contend training is an in-between process that captures statistical regularities rather than human expression.
i. The American Framework: The Fair Use Defense
Litigation is playing out in US courts on the issue of 107 of the Copyright Act-Fair Use in particular. AI companies argue that model training is a profoundly “transformative” use because it employs copyrighted expression as a way of building a statistical reasoning engine rather than copying and releasing the actual works.
But creators and other media companies are challenging this, saying that training the AI itself actually undercuts the market for its creations. When a generative model generates something that directly replaces the journalism, artwork, or computer code that would have trained it in the first place, the non-competing transformative use argument crumbles.
ii. The Indian Framework: Fair Dealing and Section 52
Fair use in India Despite the flexible four-factor fair use doctrine in the US, India’s Copyright Act, 1957 only enumerates fair dealing exceptions in Section 52. It is important to note that Section 52(1)(a) expressly provides a fair dealing exemption for private or personal use, including “research”.
The first significant challenge to this statutory doctrine came before the Delhi High Court in ANI Media Pvt. Ltd. V. OpenAI Inc. (2026). ANI claimed its copyrighted news reporting was used to train ChatGPT without permission or remuneration. The Court refused to allow an interim injunction, opining on a prima facie basis that model training and temporary storage of content for machine learning was covered by the research exception of Section 52(1)(a).
The Court also recognized some wider notions of public interest that technology can bring. But, as Section 52 is silent on a free-and-easy ‘transformative use’ defense, this leaves Indian Courts with difficult issues: whether profiteering training of AI could ever – in future – be categorized under existing exceptions to copyright.
2. The Output Frontier: The Imperative of Human Authorship
The legal grounds for input training are still under dispute, but the international position on output is much less ambiguous: solely AI-created work is outside the scope of copyright.
i. The Human Centricity of Copyright
Copyright law has a noble objective to encourage human mental labor and intellectual expression. This noble objective is universally favored:
a. United States: The US Copyright Office and federal courts have recognized a requirement that copyright be created by a human author. The submission of natural language prompts to a generative model does not identify the user as an “author” because the prompt is directed at a machine that produces its own expression.
b. European Union: EU case law considers a work to be eligible for protection only where it is that of “the author’s own intellectual creation”, which presumes a human element infused into the work.
c. India: Under Indian Copyright Act (section 2 (d)), an author with respect to a work which is made by a computer: “he is the person by whom the work is done.” Judicial interpretation has added, “in which human skill, judgment and labour are directly involved in the final expression.”
ii. The Hybrid Frontier
A second approach occurs when human authors embrace generative AI as a supplemental resource: They start with an AI generation and then shape and arrange it, editorially, into a new derivative work. In this hybrid approach, only the human-created material receives copyright protection, and the AI component remains free for all to use.
Structural Innovations: How Copyright Frameworks Are Adapting
Instead of being crushed by generative technology, copyright law is being reshaped by regulatory reform, technical measures for enforcement, and market forces.
1. The European Union’s AI Act: Statutory Transparency
The EU AI Act contains mandatory legal requirements regarding copyright transparency for providers of GPAI models. These are as follows, according to Article 53 of the Act:
- Keep in-depth technical records about the origin of the training data.
- Reproduce at a granular level a reasonably detailed summary of copyrighted works used to train their models.
- Use technical measures to implement opt-out reservations by rightsholders made under Article 4(3) of the Digital Single Market Directive (e.g. Robots.txt flags, automated machine-readable metadata, etc).
” Failing to comply with the standard will lead to substantial administrative penalties – setting a global benchmark of enforcement so that technology providers have to be mindful of the origin of their training data.”
2. The Shift Toward Voluntary Collective Licensing
Aware of the risks for long and uncertain litigation, technology companies and content owners are turning more and more to collective licensing. Leading news publishers, stock photo collections and recording labels are striking licensing agreements with AI developers. Such licenses give developers legal clarity and high-quality data pipelines, while ensuring a continuing income stream for human creators.
3. Technical Provenance and Copyright Management Information (CMI)
Rights enforcement is also moving to technical standards within an automated digital environment. Standards such as C2PA (Coalition for Content Provenance and Authenticity) embed cryptographic data into content to reveal authorship, edits and training preferences. Under the US Digital Millennium Copyright Act (DMCA) Section 1202, removing or modifying that CMI when data is scraped also creates independent liability for AI developers.
Comparative Legal Analysis across Epochs
The generative ai era therefore takes a fundamental turn away from traditional copyright doctrine in 4 directions:
Initially, the underlying safeguarding focus has moved from static, fixed expressive works – physical books, movies, drawings etc. – and towards the validation of real human provenance, curation, and authorship information.
Second, regimes for ingestion have shifted from strictly permitted copying with known license upfront to a playing field of legal tension between Fair Use, the statutory Fair Dealing research exceptions and required collective licensing.
Third, statutory authorship rules are changing. Where purely human-created work was the only option under traditional law, today’s rules, taking into account many outside factors, place AI-only created work in the public domain and offer limited protection to blended works created with significant guidance and editing from humans.
Fourth, the regulatory enforcement strategy is moving from retroactive litigation against illicit copying to a preemptive imposition of technological interoperability (machine-readable opt-outs, embedded Copyright Management Information [CMI]).
Conclusion
Copyright law can-and will-endure generative AI, but it will not endure in its analogue version. By virtue of its inherent squabble with generative machine learning and intellectual property, the physical copying age in which we live is coming to an end-the coming age is where we place our value on human authenticity.
In the past, copyright was based on the idea that reproducing creative work was costly, complex, and time-consuming, but that it was also easy to recognize. Digital copying is quick, cheap, and easy to conceal, and Generative AI creates its synthesised content instantaneously. If copyright law continues to only prohibit storing or intermediary statistical analysis of digital files, it will hinder innovations while offering no protection for creators.
Rather, the modern doctrine needs to be refocused to three primary purposes:
1. Crack Down on Ingestion Transparency:
Regulators should put a stop to the ‘black box’ operation of AI developers. Requiring public, standard formats for high-trust auditing of training datasets can make sure right’s holders aren’t being exploited to make money.
2. Maintaining Licensing Markets:
Courts and lawmakers will need to safeguard original authorship by prohibiting commercial AIs from crawling human works to produce direct market substitutes without some form of compensatory purchase or contractual opt-out.
3. Public Domain and Human Authorship:
Protecting public domain content for the human author defends against corporate crowding out of automated media with artificially cheap, monopolized media.
Copyright law has always been about human rights: protecting our dignity, our work, our stories, our expression. Courts and lawmakers can safeguard copyright’s human core in AI era by drawing lines around training transparency, honoring technical opt out, and prioritizing human authorship. This will preserve copyright as a vital institution for human culture, not a resource for the machines.
References
2. Regulation (EU) 2024/1689 (European Union Artificial Intelligence Act): Codifying mandatory copyright transparency obligations, data provenance documentation, and text-and-data mining (TDM) opt-out enforcement under Articles 53 and 54.
3. Bartz v. Anthropic PBC & Kadrey v. Meta Platforms, Inc. (US District Courts): High-profile US litigation examining whether large-scale commercial ingestion of copyrighted books for LLM training constitutes transformative Fair Use.
4. Justice K.S. Puttaswamy (Retd.) v. Union of India, (2017) 10 SCC 1: Foundational Supreme Court of India precedent establishing the primacy of individual dignity and autonomy in digital systems.
5. Thaler v. Perlmutter, 104 F.4th 912 (D.C. Cir. 2024): Affirmed the fundamental requirement of human authorship for copyright protection under modern intellectual property law.
