Generative AI, AI Authorship, Technological Generator, Human Authorship, Copyright

Introduction : Imagine that you receive a new client who seeks to bring an action for copyright infringement. As you learn more about the case, you realise that your client is ‘morally’ aggrieved or rather, ‘algorithmically disappointed’ and legally injured because a work it created after hours of painstaking neural-network processes involving training, association formation, perturbation, monitoring, stabilisation, and output generation has been infringed. Sounds absurd? Perhaps. But considering all the buzz around AI today, in five years, you may just find yourself standing in court, arguing for your very own ‘artificially intelligent’ client.

On August 31, the registrar of copyrights in the application of Dr. Stephen Thaler to name his AI system DABUS (Device for the Autonomous Bootstrapping of Unified Sentience) as an author of the artwork ‘A Recent Entrance to Paradise’, it held that the AI-generated work satisfies the originality requirement under Section 13 by meeting the minimum degree of originality and the threshold of being independently created rather than copied, although the model itself is a computational AI system. Even though the originality of the work is agreed upon, the Copyright office held that AI cannot be considered as an author under Section 2(d)(vi). The decision is premised on the fact that AI is not a natural or Juristic person recognised in Law. What needs to be considered is the possibility of a new category of ‘Technological generator’ to be created exclusively for accommodating such systems.

The Order

What makes the registrar’s order worth considering for a close reading is the reasoning and the underlying question it seeks to address. The question wasn’t merely who performed the mechanical act of generating the work. Instead, the Registrar asked who was the “mastermind” behind it , echoing an idea the US Copyright Office had itself invoked three and a half years earlier, although in a completely different context.

The reasoning then moves towards the idea of the “effective cause” of creation. Drawing from Burrow-Giles Lithographic Co. v. Sarony and Aalmuhammed v. Lee, and reviewing the structure of Section 2(d) itself, the Registrar distinguished between the person who actually operates the mechanism and the person whose intervention sets the creative process in motion. After all, the Act attributes authorship of a film to its producer, rather than its cameraman, and of a photograph to the person who takes it, rather than the person who built the camera.

Applying this logic to DABUS, the Registrar found that Thaler was the person who conceived and configured the system and supplied the inputs that set the process in motion. He was therefore the person who “caused” the work to be created, while DABUS merely executed that cause. There is, however, an interesting twist. Thaler had been given opportunities across three hearings to amend the application and name himself as the author, but refused to do so. He continued to insist that DABUS itself be recorded as the author. The application therefore failed on the very basis on which it was presented, rather than being modified into a version that could potentially succeed.

But the order makes another distinction that may be even more important for the future of AI and copyright: originality and authorship are not the same inquiry. Even if the last generative step is done by the AI model without a human prompt, the originality threshold that Section 13 requires is slightly higher. It needs the works to be created independently rather than a mere mechanical production of already existing work.  

Therefore, according to the registrar’s reasoning, an AI model can create original works but cannot be an author. That slot remains reserved for the person who stands behind the creation: the ‘mastermind’ who, in the eyes of the law, caused the work to come into existence.

When the Software shared the Byline

Six years before DABUS, India had already encountered a smaller, stranger version of the same question, and never quite finished answering it. In November 2020, the Indian Copyright Office registered Suryast: a sunset photograph taken by IP lawyer Ankit Sahni, processed through an AI painting application called RAGHAV and stylised in the manner of van Gogh’s The Starry Night. The registration  was unusual as Sahni and RAGHAV themselves were listed as co-authors. By most accounts, it was the first time a copyright register had named software as a joint author.

A year later, however, the story became less clear. Around 25 November 2021, the Copyright Office reportedly issued a notice seeking to withdraw the registration of  co-authorship. Yet the fullest independent account of the dispute states that the Registry “refused to take steps to correct” the registration further and that, despite the notice, it has never been formally expunged.

There is still some uncertainty: an account relayed through Sahni suggests that the withdrawal notice itself was later withdrawn. If so, the world's first AI co-authorship registration may technically still be sitting on India's Register of Copyrights.

Does DABUS Settle Suryast?

The DABUS issue is concerning sole authorship to be granted to AI, while Suryast is about joint authorship between a human and an AI system. Section 2(z) The Copyright Act defines joint authorship as a work “produced by the collaboration of two or more authors in which the contribution of one author is not distinct from the contribution of the other”. It assumes multiple authors who are capable of collaborating. The threshold of authorship under the joint contribution shall be independently satisfied by all the authors. What DABUS asks is not only whether AI can be an author but also whether India's Copyright Office has any mechanism, short of the pending Canadian-style litigation, to revisit a registration whose own later reasoning now appears to contradict.

The question asked by Suryast, is a different one that, can a human and an AI share the authorship, even though the latter is not a legally recognised person. It's even more interesting when their treatment in different jurisdictions is also considered. The same problem has been dealt with differently by different legal responses. Taken. The US Supreme Court kept its opinion towards human authorship by not disturbing the decision by the DC circuit in favour of the human authorship requirement in Thaler v. Perlmutter.

India's DABUS order, under a very different statutory framework, reached much the same destination. The vocabulary differed “human authorship” in the United States, “mastermind” and “the person who causes the work to be created” in India, but the underlying conclusion was similar: the machine may generate the work, but the law still looks for a human being behind it.

The picture changes for co-authorship. India granted the registration, Canada granted an equivalent registration in December 2021, while the United States rejected the claim twice. In its December 2023 reconsideration, the US Copyright Review Board found Sahni's contribution insufficient for human authorship and concluded that RAGHAV was responsible for the work's final expressive form.

Canada’s registration is now being directly challenged in the court. The Samuelson-Glushko Canadian Internet Policy and Public Interest Clinic, filed an application before the Federal Court, In July 2024, seeking to expunge or rectify the Suryast registration, on the basis that “an author must be a human being” and, alternatively, that Sahni alone should be named as its author.

Suryast, therefore, leaves us with a question DABUS did not answer: if AI cannot be an author on its own, can it nevertheless share the byline with a human? None of this difference in decisions is accidental. Each jurisdiction is applying different statutory raw material to essentially the same problem, and the difference in text explains much of the difference in outcome.

India expressly provides that, for a computer-generated work, the author is “the person who causes the work to be created.” The UK takes a more direct route “Where there is no human author, the author is deemed to be the person who undertakes the arrangements necessary for creating the work”. The United States has no equivalent statutory provision. Section 102(a) simply protects “original works of authorship”; the human-authorship requirement has instead been developed through judicial interpretation.Canada goes further in the opposite direction: it does not provide a statutory definition of “author” at all, leaving precisely that question to judicial interpretation in the pending Suryast litigation.

Put together, the four jurisdictions reveal four different approaches to the same problem: India stretches a human-centered provision to accommodate computer-generated works; the UK has a specific statutory fiction; the US has built its position through case law; and Canada is still working out the answer.

The Originality

But before asking who the author is, there is an older question to answer: is the work original in the first place? India's Copyright Act does not define “originality”; its meaning has largely been developed through case law. Interestingly, the standards across jurisdictions are not identical. India and Canada follow a skill and judgment approach, requiring something more than trivial or purely mechanical effort, but not necessarily creativity in the strict sense, as seen in Eastern Book Co. v. D.B. Modak and CCH Canadian Ltd. v. Law Society of Upper Canada. The United States, following Feist Publications, Inc. v. Rural Telephone Service Co., requires independent creation together with a minimal degree of creativity. The UK and European Union, meanwhile, focus on the work being “the author's own intellectual creation, involving free and creative choices”, as reflected in THJ Systems Ltd. v. Sheridan and Infopaq International A/S v. Danske Dagblades Forening.

The doctrinal journey appears itself revealing. The older English “sweat of the brow” approach developed into the “skill, labour and judgment” test, which India continued to apply before its Supreme Court moved towards the Canadian approach in Eastern Book Co. The US Supreme Court, meanwhile, rejected “sweat of the brow” in Feist and insisted on at least a “modicum of creativity.” The European Union developed its own “author's own intellectual creation” standard through Infopaq.

No two jurisdictions therefore apply precisely the same test. Yet India's and Canada's approaches, skill and judgment, instead of purely mechanical effort, without demanding a high degree of creativity, remain strikingly close.

It is tempting to draw a line from that shared originality standard to a related coincidence. India and Canada are also the two jurisdictions in this story where an AI's name actually made it onto a copyright register. The line should be drawn carefully, though. Originality and authorship are, as already noted, separate inquiries, and Canada's registration is a weak data point for any claim about its originality test specifically: CIPO's registration process is unexamined and automated, so Suryast was never actually tested against the skill-and-judgment standard, or against any authorship standard, before it reached the register, a gap CIPPIC itself relies on in arguing the registration should be expunged. A narrower one is more defensible, it is the substantive, judicially developed approach to originality in India which left room for the DABUS work to qualify; whereas the registration in Canada reflects an administrative gap rather than a considered application of the same test.

Suryast's status in India remains ambiguous; CIPPIC v. Sahni is still pending before Canada's Federal Court; and, most importantly, the DABUS order deliberately left the larger question of AI legal personhood to Parliament.

Running beneath the authorship debate, is another copyright fight: not who authored a work, but whether AI's ingestion of copyrighted works during training is itself infringement. India's position so far comes from an interim order by the Delhi High Court in ANI Media v. OpenAI, where the Court considered whether internally held training material could fall within “private or personal use, including research.”

By contrast, Germany’s GEMA v. OpenAI drew a different line, ruling that the EU’s text-and-data-mining exception does not cover reproductions of memorized lyrics appearing in chatbot outputs.

The New York Times litigation against OpenAI and Microsoft raises the same underlying concern: whether large models memorise licensed material and can reproduce it when prompted. With that question still unresolved, AI training and AI authorship increasingly look like two sides of the same problem, not simply could a machine create?, but where does the machine's contribution begin and human legal responsibility end?

Borrowing the Author’s clothes

It is tempting to solve the problem by giving AI the legal clothes of an author. The trouble is that they do not fit very well. Generative AI complicates the traditional connection between an author's personality and the work: its outputs emerge through “algorithmically mediated statistical associations” rather than a straightforward act of human creative will.

That problem becomes sharper when we move beyond originality to the rest of the copyright bundle. Apart from the liability gap, the effectiveness of awarding AI with authorship credits at the same time retaining human centered characteristics of authorship like moral rights, honour, reputation, responsibility, fairness, and when it comes to the infringement, intention and knowledge, questions the very intention of doing so. 

The European Parliament’s earlier proposal of extending legal status to robots, or in other words, electronic personhood, was widely criticised by the experts, citing the same concerns.  Even the corporate person analogy cannot be taken without a pinch of salt, as they were given rights and liabilities owing to the fact that they are run by the actions of real people.” The plan was later dropped in the EU AI legislation. The DABUS cases also point in the same direction, as the decision rejects the proposition that an AI system can itself be an inventor.

But rejecting AI legal personhood cannot bring an end to the discussion. There are several serious questions posed by scholars worldwide regarding different aspects of  AI Authorship.

Annemarie Bridy suggests a model where the programmer or human behind the system as author-in-law, to circumvent the need to vest rights in a machine, preserving the incentive structure of copyright. Similarly,  Ryan Abbott argues that, even if the creator is a machine, providing incentives to the human or corporate developers can encourage creative production. At the same time, Francesca Mazzi says that even where the final expression is generated by the  AI model, if the human prompt is sufficiently deliberate and it may embody the creator’s unique creative intent and personality”. Now, imagine the situation where there is no prompt behind the final output, as Thaler claims. The most practical middle ground can be seen in the US Copyright Office’s Zarya of the Dawn decision.  It protected the author's written text and her selection and arrangement of the AI-generated images, but not the individual Midjourney images themselves. The Office reasoned that prompting and refining a prompt did not mean the user had actually formed the resulting image.

 Interestingly, the US Copyright Office reached for the same idea in 2023, asking whether the human prompter was in substance the “master mind” behind a Midjourney image, and India's Registrar closed the loop years later by asking who the “mastermind” behind DABUS's output was. It poses  the same underlying question: whose will, exactly, is this work an expression of? Two artworks, four countries, six years and, as of this writing, still no single, universally accepted answer. But look closely at where each path ultimately terminates, rather than where it merely pauses, and a pattern emerges. The US Supreme Court declined to disturb the D.C. Circuit's human-authorship ruling. India's Registrar, applying a sixty-year-old statute never drafted with generative software in mind, arrived at it independently. The UK wrote it into statute outright, decades before anyone had heard the phrase “large language model.” Even the European Parliament's attempt to legislate around it collapsed under the burden of more than 150 scholars pointing to the same structural gap.

The pattern is this: every one of these systems, whatever else divides them, insists that behind the work there is someone, a mastermind, a person who causes the work to be created, a person by whom the arrangements necessary for its creation are undertaken, and not merely something.

Conclusion

The question of rights and duties brings us back to a legally recognized person, whether AI is ultimately granted this status or not. At present, the Indian framework does not recognize AI as a legal personhood and, as such, AI is unable to, independently, possess the rights or incur the liabilities that flow from authorship. Section 2(d)(vi) of the Copyright Act, 1957 deals with computer generated works, by defining the author as ‘the person who causes the work to be created’.The DABUS controversy shows the tensions of forcing increasingly autonomous AI into a legal framework built on the idea of a human ‘person’ behind the creation. But, the rights, actions and  liabilities linked to the copyright cannot be easily attached to the technological entity.

Solution here shall not be merely providing an equal status for AI models like that of the human authors.  A more pragmatic approach could be to embed AI as a tool, technological system or distinct category of technological generator in existing legal structures without prematurely ascribing to it every legal attribute of a human being.

The question here is not simply whether AI should be called an ‘author’, but whether the concept of authorship as it stands today is flexible enough to accommodate highly automated technologies with generative capabilities.

This is where the concept of giving AI the identity of a ‘technological generator’ becomes interesting. Rather than immediately granting AI statutory recognition, proprietary status or legal personality, the law could recognise its technological role in generating a work while defining the rights and limits that follow. This could provide greater transparency around AI-generated works without prematurely deciding the question of legal personhood. This is not an unprecedented legal framework, the US copyright office's approach in Zarya of the Dawn, protecting the human's selection and arrangement while declining to protect the AI-generated images themselves, already discusses the idea. Technological generator status would be helpful to formalise such arrangements as a general rule rather than a case by case workaround. 

At the same time, as AI's effect reaches beyond authorship to training on copyrighted works, research exceptions, fair dealing, ownership, attribution, infringement and liability and Fair dealing being already recognised in section 52 of the Copyright Act for purposes including research, criticism and review, the challenge is working out how these principles apply to increasingly sophisticated artificial intelligence (AI) systems. Thus, the goal may not be to make AI human in the eyes of the law, but to make the law flexible enough to take on AI without losing its human-centred foundation.

And yet, I cannot help but wonder whether I will cite an AI author in my next paper and give it credit in the footnote as ‘AI et al.’?

Author: Gauri Parvathy R in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at  Khurana & Khurana, Advocates and IP Attorney