Abstract
Generative artificial intelligence has fractured one of copyright law’s oldest assumptions that authors are always human. Navigating statutes, judicial decisions and copyright office guidance, this article argues that no jurisdiction has yet produced a workable answer to who owns AI generated work. Indian brands have deployed AI generated celebrity images in advertising without licenses and AI news anchors broadcast on national television channels, Stability AI trained its image model on Getty’s licensed photographs without permission. The article concludes that incremental judicial interpretation is not equal to the problem and makes the case for sui generis legislative reform.
I. Introduction
Copyright law has survived a number of clashes with technology. Photography, sound recording, digital copying each forced court and legislature to stretch existing categories in ways their drafters had not expected. The problem of AI generation is not the presence of a new medium, but a challenge to the idea that authorship necessarily means human authorship.
When someone writes a prompt to Midjourney and gets a finished output. When a agency has Runway make a 30 second ad that involved no human animators. It doesn’t fit anywhere in copyright law. No human programmer actually intended for this specific result. The user who typed a sentence of instruction contributed something but whether that something amounts to the kind of creative expression copyright law has historically protected is a genuinely open question. So who, if anyone owns the work?
In 2022, Aaj Tak debuted Sana, an AI news anchor with the voice and likeness of a real journalist that broadcast on national television. There is still no clarity on copyright of her broadcasts or rights of the individual whose biometric data she simulates. Indian advertising companies like dentsu India are already using AI generated visuals in campaigns without clarity on copyright of such visuals or licensed use of training data. Internationally, Getty images filed a suit against Stability AI in the English High Court and the New York Times sued OpenAI in Manhattan federal court alleging that their copyrighted works had been ingested into training datasets without permission. The law has not kept pace.
II. From Tool to Creator : The Evolution of AI Generated Content
Early AI tools like software that auto generated financial reports or completed melodic phrases according to preset rules could reasonably be treated as sophisticated instruments. The human made the meaningful creative choices, the software executed them but that metaphor doesn’t scale to large language models and diffusion image generation.
Because models like GPT 4, DALL E 3, Midjourney and Stable Diffusion are all trained on large datasets of already existing human output and rather than executing explicit user content instructions, they learn probabilities of possible output based on billions of learned associations. The human contribution is real, the prompt shapes the output but it is attenuated and it is not the kind of expressive control over form and content that copyright doctrine has treated as sufficient for authorship.
In India, AI generated devotional music remixes, some featuring synthesised versions of the voices of deceased or living classical musicians circulating on Youtube and Instagram with millions of views have proliferated without any copyright enforcement mechanism being applied. There has been bollywood style AI generated advertisement posters without giving credits to artists or studios. Internationally, Drake and Taylor Swift voiced AI generated music went viral with millions of views on the streaming platforms before being taken down. It proves that neither platform content moderation nor copyright mechanisms are adjusted to the pace of content generation.
III. The Human at the Heart of the Statute : The Human Authorship Requirement
The requirement that an author be human runs through the structure of copyright statutes even where not made explicit.In India, Section 2 (d) of the Copyright Act 1957 defines author through enumerated categories like the composer of a literary work, the artist, the photographer, the producer of a cinematograph film, all presupposing natural or juristic persons. The 2012 amendments introduced no provision for non human authorship.
In Feist Publication v. Rural Telephone Service Co, the US Supreme Court held that copyright requires a minimal degree of creative expression, not mere labour. In Naruto v. Slater, the Ninth Circuit held that a macaque monkey could not hold copyright because the term authors does not extend to animals reasoning that applies equally to AI. The US Copyright Office confirmed in its 2023 guidance that purely AI generated works will not be registered. In Thaler v. Perlmutter the DC District Court upheld that position holding human authorship a constitutional and statutory prerequisite.
Section 9(3) of the UK Copyright, Designs and Patents Act 1988 provides that for work that is computer generated, the author is the person who undertakes the arrangements necessary for its creation, a provision most scholars regard as ill suited to generative AI. The EU standard is more demanding still, after Infopaq International V. Danske Dagblades Forening and Painer v. Standard VerlagsGmbH, a protected work must reflect the author’s own intellectual creation expressing their personality through free and creative choices. Statistical inference over training data cannot satisfy this.
IV. Everybody’s Work, Nobody’s Right : Competing Claims in AI Generated Content
Even where some human contribution can be identified, ownership does not resolve itself. There are at least 4 plausible claimants including the developer, the user who provided the prompt, the authors of works used in training and the public. The developers claim rests on work for hire analogy but copyright has never protected the maker of an instrument as the author of what it produces. Ryan Abbott has argued that a user who provides a highly specific prompt exercises sufficient creative direction to qualify as its author and this has some force but it requires copyright law to treat the arrangement of high level instructions as equivalent to the arrangement of expressive content, an equation the US Copyright Office has declined to make.
The training data question is the most commercially consequential. Getty Images litigation against Stability AI puts directly in issue whether a model trained on copyrighted images without a license infringes those copyrights either through the training process or through outputs that reproduce composition or distinctive elements. If training is infringing, the economic model underlying most commercially deployed generative AI is legally precarious.
In India, when an AI tool trained on Carnatic classical recordings generates a devotional composition circulating on Youtube under a user’s name, multiple interests are simultaneously in play like composer’s rights in the underlying musical works, performers rights under Section 38 in their recorded performances and the personality rights of living artists whose vocal style has been replicated.
Some Indian brands used deepfakes/ AI generated images of living celebrities in their advertisements in 2023 but did not have rights to use them. The doctrinal tools available like passing off, the right of publicity recognised in ICC Development (International) Ltd. v. Arvee Enterprises, moral rights under Section 57 were developed for factual contexts quite different from synthetic voice and likeness replication. The Copyright Act does not directly protect a performers voice where it has been synthetically reproduced rather than recorded.
One example is Cadbury India’s 2023 Diwali Campaign which used AI to recreate Shah Rukh Khan’s voice and likeness without a fresh license for each regional variation. According to 38A, a performer has rights over the reproduction of their recorded performance and whether an AI generated reconstruction constitutes reproduction within that provision has no judicial answer. Cadbury example is in the contractual grey area where AI gives the brand a greater benefit and more commercial value from a licensethan the parties had ever envisioned.
V. Comparative Jurisdictional Analysis : Four Jurisdictions, Four Gambles
The US position takes human creativity as the constitutional requirement for copyright. The UK position under Section 9(3) always results in the granting of a rights holder but completely disconnects authorship from creativity.The EU position as in no expression of human personality no protection is philosophically consistent. India has the most unclear position as its Copyright act almost definitely does not grant protection to purely AIgenerated work. Whether training on Indian copyrighted content constitutes infringement under Section 51 and how performer protections under Sections 38 and 38A apply to AI synthesised vocal output have no statutory answer.
VI. Major Legal and Policy Challenges : The Cracks Beneath the Surface
In the short term there will be 3 issues driving litigation and reform, the first is the training data problem. The New York Times suit against OpenAI and Microsoft filed in December 2023 alleges GPT 4 was trained on millions of Times articles without a license. In India, the equivalent question falls under Section 52 of the Copyright Act which sets out a narrower permitted acts regime than US fair use and includes no general transformative use exception covering commercial AI training.
The second is moral rights. Section 57 of the Indian Copyright Act gives authors the right to claim authorship and to object to modifications prejudicing their honour or reputation, rights surviving the transfer of economic rights. When a generative AI system produces stylistically derivative outputs from a human authors work, whether those moral rights are engaged is not obvious.
The third is platform liability. Section 79 of the Information Technology Act 2000 and the Intermediary Guidelines of 2021 provide a safe harbour for platforms hosting third party content built for hosts, not for platforms deploying their own AI tools. When a platform is itself the generator, the safe harbour logic does not apply.
VII. Beyond Patchwork : Need for Reform and Suggested Solutions
No workable legal solution will arise from judicial intervention alone. The courts may determine whether AI created works should not be the subject of copyright but they are incapable of devising the positive legal instruments, namely a system for licensing training data, the creation of a sui generis intellectual property right and the introduction of a new personality rights statute.
The Copyright Act could be amended to differentiate between truly AI created works and AI assisted works, the latter having a human author that exercised a specified level of originality in respect of which copyright subsists. For the issue of training data, a compulsory licensing regime on the lines of the compulsory license for cable retransmission that requires the disclose of training corpora and the contribution of sums to a central royalty pot is more in keeping with copyright principles than declaring that training is not inherently infringing as both Ginsburg and Budiardjo have argued. On personality rights, the right of publicity from ICC v. Arvee Enterprises, performers’ rights under Section 38A and passing off are 3 inadequate partial answers requiring one complete solution. A dedicated Right of Publicity Act adapted to Indian Constitutional values around dignity and identity would close this gap more reliably than continued common law improvisation.Mandatory disclosure of AI generated content in commercial, journalistic and broadcast contexts is long overdue, the EU AI Act’s transparency provisions offer a workable model.
VIII. Conclusion
Copyright laws encounter with generative AI is not a standard technology disruption story. The human authorship rule is also based on a commitment regarding law and creative personhood. An abandonment entirely of human authorship would redefine copyright law moving it from a law that rewards human creativity into one that protects capital investment in machine infrastructure. Equally a framework that denies protection to all AIgenerated content and treats training on human authored work as free leaves human creators with no remedy and no share of the commercial value their work has generated. The legal vacum in india does not prevent exploitation it just means the people being exploited haveno recourse.
THIS ARTICLE IS WRITTEN BY PRISHA CHAUDHRY FROM O.P. JINDAL GLOBAL UNIVERSITY / JINDAL GLOBAL LAW SCHOOL
REFERENCE :
The Copyright Act, 1957, s. 2(d). The provision defines ‘author’ through an exhaustive list of creator categories, all of which presuppose natural or juristic persons. No provision in the Act addresses non-human authorship, and the 2012 amendments did not introduce one.
Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991). The Court rejected the ‘sweat of the brow’ doctrine and held that copyright subsists only in original works embodying a minimal degree of creativity, not mere labour or investment.
Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018). The Ninth Circuit held the Copyright Act’s authorship provisions do not extend to non-human animals.
U.S. Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16190 (16 March 2023); Copyright and Artificial Intelligence, Part 1: Digital Replicas (2023).
Thaler v. Perlmutter, No. 22-cv-1564 (D.D.C. 18 August 2023). The court affirmed the Copyright Office’s refusal to register ‘A Recent Entrance to Paradise,’ holding human authorship to be a constitutional and statutory prerequisite for copyright protection in the United States.
Copyright, Designs and Patents Act 1988 (UK), ss. 9(3), 178. Section 9(3): ‘In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.’
Infopaq International A/S v. Danske Dagblades Forening, Case C-5/08, [2009] ECR I-6569 (CJEU); Eva-Maria Painer v. Standard VerlagsGmbH, Case C-145/10 [2012] ECDR 6 (CJEU).
Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge University Press 2020) 50–74. Abbott argues for a user-centred authorship model based on the quality and specificity of the creative direction exercised through prompting.
Shyamkrishna Balganesh, ‘The Pragmatic Incrementalism of Common Law Intellectual Property’ (2008) 63 Vanderbilt Law Review 1543; Ananth Padmanabhan, Intellectual Property Rights: Infringement and Remedies (LexisNexis 2012) ch 9. Both advocate for a dedicated publicity rights statute in India adapted to its constitutional framework.
EU AI Act (n 13) Art. 50. The provision requires deployers of AI systems generating synthetic audio, image, video, or text content to ensure outputs are marked as artificially generated in a machine-readable format.
