As the Delhi High Court directs the Registrar of Copyright to consider Stephen Thaler’s bid to register a fully AI-generated artwork under Indian copyright law, a deeper question emerges: was the Copyright Act ever designed to recognise non-human authorship at all? In light of this development, Rashi Singhal argues that the architecture, rationale, and core provisions of the Act remain fundamentally human-centric, making fully autonomous AI authorship difficult to accommodate within the existing framework. Rashi is a Graduate Research Fellow at the JSW Centre for the Future of Law, National Law School of India University (NLSIU), Bengaluru.
Can the Indian Copyright Act Accommodate Fully Autonomous AI Generated Output?
By Rashi Singhal
On April 9, 2026, the Delhi High Court directed the Registrar of Copyrights to rule on Stephen Thaler’s pending application (Diary No. 9356/2022-CO/A) within eight weeks of its April 27 hearing. This application, filed in 2022, seeks copyright registration of A Recent Entrance to Paradise, an artwork autonomously generated by AI. A similar application by Thaler has been rejected in the US, with the Supreme Court denying certiorari in March 2026, on the ground that US copyright law requires human authorship.
The AI output copyright question is not one question but a set of questions along a spectrum. At one end are cases such as Thaler, where the applicant claims the AI generated the work without traditional human authorship, sometimes claimed to be “fully autonomous” generation. At the other end are cases where AI is used purely as an assistive tool, where the human author is not displaced. In between lie contested cases where a user iterates through dozens of prompts and selects carefully, and the question is whether their contribution or the model’s contribution dominates, with much of the debate turning on the degree of human control over the expressive elements of the output.
This piece deals only with the “fully autonomous end,” cases where the applicant claims that AI produced the work without meaningful human creative input, and argues that the Indian Copyright Act, 1957, read as a whole, is structured around a human author and cannot be read to accommodate fully AI-generated output.
Rationale of the Act
Delhi High Court in the Rameshwari Photocopy case articulated the rationale as, “Copyright…is designed rather to stimulate activity and progress in the arts for the intellectual enrichment of the public… It is intended to motivate the creative activity of authors and inventors in order to benefit the public.”
This utilitarian theory approach is reflected in the architecture of the Act. Term limits, compulsory licensing, statutory licensing under Section 31D, and the exceptions under Section 52, are all designed to balance incentives for creators against public access. Would we need to incentivise AI to keep creating autonomously generated output? Moreover, the Act also reflects some influence of personality theory through Section 57, which treats the work as an extension of the author and recognises “moral rights.” These rights protect the author’s identification with the work and the work as an expression of the author’s personality. Together with the utilitarian incentive structure, this points to a framework that is human-centric in its architecture – in its terms, exceptions and rationale, and not only in the provisions that define authorship.
Section 16 and 2(d)
Section 16 of the Act provides that “No person shall be entitled to copyright … otherwise than under and in accordance with the provisions of this Act …” Copyright in India is a statutory right, and the rights exist only where the statute creates them. This is confirmed by the pattern of amendments: each time Parliament has extended protection to new categories of works, such as computer programs, computer-generated works or performers’ rights, it has done so by amending the Act.
Section 16 places a threshold requirement that the rights-holder be a person. “Person,” under Section 3(42) of the General Clauses Act, 1897, includes “any company or association or body of individuals, whether incorporated or not,” extending the meaning to natural persons and juristic persons. However, it does not extend to entities, such as AI, that lack legal personality altogether.
Section 2(d) defines “author” by category, using terms that refer to the human role in each type of work. As originally drafted, it assumed that authors were humans engaged in these specific creative activities.
The 1994 amendment added Section 2(d)(vi): “…work which is computer-generated, the person who causes the work to be created.” The question remains whether this provision could be read to argue for the accommodation of AI as the author. I argue that it cannot.
The first thing to notice is that the provision exists at all. Every carve-out in the statute proves the existence of a rule it departs from. The second is that this provision designates a person as the author. It does not deal with non-human authorship, rather assumes a human whose causal contribution is sufficient to make them the author. Additionally, the provision was added in 1994, when the computer-generated works Parliament had in mind were works produced using computers as tools, not the autonomous generation in question today. Reading Section 2(d)(vi) to accommodate fully autonomous AI output is anachronistic.
Section 13(1)
As per Section 13(1), the originality requirement applies only to works under Section 13(1)(a). In India, originality can be understood as: the work originates from the author and reflects “skill and judgment.” The first step itself has two parts: the work is not copied, and it originates from the author. The second step requires the exercise of “skill and judgment” that is “not the product of merely labour and capital” and not “so trivial that it could be characterized as a purely mechanical exercise.”
One way to check whether there is originality in the AI output is to run the test on the output. I propose that this test cannot be run on the output alone, as the test asks a question about the relation between an author and the work the author produced, such as, did this author originate this work, and did this author exercise “skill and judgment” in producing it? The question can only be answered if there is an author whose relation to the work is being assessed.
For human authors, the test works because the author is given. The origination question is proved by the work not being a copy, along with the natural assumption that it originated from the author. For “skill and judgement,” one can look at what the author did, examine the creative process, and decide whether her relation to the work satisfies the test.
For AI output, the test fails because there is no author whose relation to the work the test can examine. One can try to locate a human, for instance, the user or the developer, but any candidate would be a reconstruction, not a bearer the test can straightforwardly assess.
The present question makes visible a baked-in assumption. “Originated from the author” is usually treated as equivalent to “not copied from another work,” if the work is not copied, then it originated with the author, because who else would it have come from? This equivalence holds for human authors because the human case forecloses any other possibility. For AI output, ruling out copying does not by itself establish that the work originated with any particular author. While the two points run together for human originated works, they come apart for AI. This shows that the originality test was designed with the assumption of a human author.
The “skill and judgment” test also does something similar. The test asks whether the author exercised “skill” and “judgement” in producing the work. For a human author, one can evaluate her skill and judgment by examining her process. For AI output, the skill and judgement the test is asking about have no identifiable bearer.
Thus, the originality test, as we understand it now, cannot coherently run on AI output.
Other Provisions
Section 14 defines copyright as the exclusive right to do or authorise certain acts. These rights presuppose an agent capable of exercising them. Sections 18 to 21 reinforce this. They deal with assignment, mode of assignment, and relinquishment, all of which require an entity capable of entering legal agreements. These are acts that can only be performed by an entity with legal personhood.
The utilitarian rationale is to give authors monopoly rights to ensure future creation, the exercise of the monopoly being how the author benefits. AI cannot exercise these rights, nor does it need this incentivisation to keep creating. Any attempt to vest copyright in AI collapses into a question about who the rights-holder actually is, and the statute assumes it is either a natural person or a juristic person.
Section 17 makes the author the first owner, subject to provisos that reallocate ownership in certain cases, showing that the Act accommodates non-human ownership when it is explicitly provided so. However, they do not accommodate non-human authorship. This distinction the Act maintains between authorship and ownership means that recognising a non-human author would require more than an expansive reading of Section 2(d)(vi). It would require revising the relation between Section 2(d) and 17.
Chapter V governs the term of copyright. Section 22 provides the default term as life plus sixty years. There is a default assumption of human life. While Sections 23 through 29 provide carve outs, they are tied to a specific category of works and to a specific reason why the default does not apply. In each case, there is still a human somewhere in the account. AI output fits none of these categories. Thus, even if the current Act could be read to accommodate AI output, Chapter V could not assign it a term. Under Section 16, copyright subsists only where the Act so provides, and this silence implies that no copyright subsists in such output.
Section 57 is grounded in personality theory, i.e., the work is an extension of the author, and the Section recognises the rights of attribution and integrity. This provision cannot be applied to AI, as it has no personality in the sense the provision contemplates, and no reputation to protect. Moreover, the 2012 amendment omitted the phrase “…done before the expiration of the term of copyright” with the effect that the right is now granted in perpetuity. In my view, perpetual moral rights should not be attached to AI. The right is perpetual because it protects an inalienable personal connection between the author and the work, a connection that holds for human authors but not for AI.
Conclusion
Applying the Act to a case of fully autonomous AI generated output requires reading multiple provisions against what they were drafted to do. Section 2(d) cannot locate an author, Section 13’s originality test has no human bearer to assess, Section 14’s rights cannot be exercised, Section 17 cannot vest ownership in an author the Act does not recognise, Chapter V cannot provide the term of protection, and Section 57 cannot attach moral rights to a personhood that is not there. The framework is being asked to do something it was not built for.
