The most important premise of modern copyright law is also its least examined, Courts today respond to artificial intelligence with a reflexive certainty: copyright protects only “human authorship”. The claim is conveyed as if it resolves the issue, as if by invoking the “human” the law can stand firm against technological progress without additional investigation, however this claim is intrinsically deceptive. The category of the “human author” has always been less stable than doctrine suggests, and the arrival of artificial intelligence does not so much challenge the rule as expose how little work it has ever done. This is most clearly articulated in Thaler v. Perlmutter [1], where it was insisted that copyright subsists only on the “fruits of human intellectual labor” appears to offer a clean boundary. Yet this formulation does not resolve the problem posed by artificial intelligence, it excludes works produced without any human involvement while leaving unaddressed the far more common scenario in which humans and machines create together. The result is a doctrine that is clear at the margins and incoherent at its core.
This incoherence is not accidental, it arises from the way copyright law has historically treated originality. In Feist Publications, Inc. v. Rural Telephone Service Co., the United States Supreme Court held that originality requires independent creation and a “modicum of creativity.[2]” Indian law followed a similar path in Eastern Book Company v. D.B. Modak, rejecting the “sweat of the brow” doctrine in favour of minimal creativity [3]. These decisions are often read as establishing a substantive threshold for protection. What is less often noticed is that they say almost nothing about authorship itself. Creativity is treated as implicitly human, not because the law has defended that proposition, but because it has never needed to. The assumption has gone untested, earlier cases reveal the same underlying logic. In Burrow-Giles Lithographic Co. v. Sarony, the Court affirmed copyright in a photograph because it represented the photographer’s “original intellectual conception. [4]” The Court affirmed copyright in a photograph because it represented the photographer’s “original intellectual conception.” The camera, though technologically sophisticated, was treated as a passive instrument. The human chose the emotive parts; the machine simply executed them. On a more fundamental level, artificial intelligence disrupts this analogy since these systems generate rather than simply record or reproduce. They use selection, variation and recombination to produce results that are not fully described by human command. Therefore, continuing to see them as tools in the same sense as cameras is a means of avoiding the limitations of current theory rather than a direct extension of it. It hides the fact that the fundamental presumptions about authorship no longer hold true while maintaining the lexicon of control.
The US Copyright Office denied the claim, and the courts upheld it, saying that the Copyright Act of 1976 requires a human author. The court identified humanity in the framework of the law, including its constraints on ownership, duration, inheritance, and intention, and ruled that authorship requires a human creator. As a result, by reinforcing a category whose coherence has never been thoroughly explored, the decision manages rather than solves the conceptual challenge faced by artificial intelligence. The US Copyright Office denied the claim, and the courts upheld it, saying that the Copyright Act of 1976 requires a human author. The court decided that authorship necessitates a human creator and recognised humanity in the legal structure, including its restrictions on ownership, duration, inheritance, and intention. Therefore, the decision manages rather than resolves the conceptual difficulty posed by artificial intelligence by reinforcing a category whose coherence has never been fully investigated. The difficulty does not lie at the extremes, but in the far more common middle ground – where human involvement is partial, iterative, and resistant to clear characterization. This is the exact point where Thaler v. Perlmutter becomes silent, it creates the appearance of resolution while leaving the most difficult concerns unsolved, enabling courts to avoid confronting the intricacies of modern creative practice.
The intuitive response is then to equate prompting with authorship. After all prompts can be detailed, stylistically rich, and expressive of human intention. Yet this move erodes a distinction that copyright law has traditionally treated as foundational. A prompt conveys what the user intends but it does not settle what the work ultimately becomes. AI systems produce outputs using probabilistic techniques that are neither completely transparent nor precisely controllable; slight modifications can result in drastically different outcomes, and the same cue can give multiple results. Under such circumstances, the human role is in no way equivalent to fixing the ultimate phrase. Therefore, acknowledging prodding as authorship would undermine a fundamental copyright concept by obfuscating the distinction between idea and expression. Although this finding is unsettling since it suggests that a large portion of what is currently referred to as “AI creativity” does not qualify for protection, discomfort does not justify giving up on doctrinal clarity.
The difficulty increases when one considers the variety of AI-assisted creativity. A completely autonomous generation with no human input is at one extreme, while traditional authorship uses AI as a smart but subservient tool. Between these poles lies a wide range of practices: users generate multiple outputs, refine prompts, select among alternatives, and sometimes substantially edit the results. It is here, in this middle ground, that copyright doctrine fails. It is unclear from the necessity for human authorship what level of involvement is necessary or what kind of involvement is crucial. It is a label rather than a test. Human needs must be transformed into a standard that can perform actual work if they are to survive. The inquiry must be whether the human exercised control over the expressive form of the work, not just whether a human was involved. This offers a more specific rule: Copyright should only exist in situations when human activity significantly affects the ultimate expressive content of the work, as opposed to only starting or influencing its creation [5]. This formulation has the advantage of clarity. It makes a distinction between contribution and control, between intention and expression. It is also in line with the fundamental logic of rulings like Burrow-Giles Lithographic Co. v. Sarony, which established authorship based on human influence over expressive outcomes.
The courts must use tangible indicators of control in order for this requirement to work, One is the degree to which the user’s input is explicit: generic or unclear prompts reveal little about authorship, whereas well-organised and thorough instructions suggest a higher degree of creative direction. Another is iterative engagement. When a user generates multiple outputs, evaluates them, and adjusts prompts appropriately, the process begins to approach curation rather than merely initiation. Direct intervention in the production itself is the most important, instead of just starting the process, actions like editing, recombination, and transformation are strong indicators that the human has engaged with expression rather than merely triggering it. However this framework encounters difficulty in borderline cases. Imagine a user who chooses one image from a hundred created outputs to be published. It’s unclear if this kind of selection qualifies as authorship. The act does not always involve shaping the expression itself, even though it may call for judgement. Selection is better regarded as a type of finding than creation when the underlying outputs are generated without much human influence. This result could appear unsatisfactory since it leaves out practices that are frequently thought of as creative. However, it still adheres to a basic tenet of copyright law: the act of creating expression is protected, not identifying or choosing from what already exists.
The Indian position illustrates both the difficulty and the opportunity of this moment. AI-generated works are not expressly covered by the Copyright Act of 1957, and court authority is limited to rulings such as Eastern Book Company v. D.B. Modak, which underlines minimum inventiveness without challenging the source. A gap is the term used to describe this silence. It would be more accurate to refer to it as a doctrinal choice space. Indian courts can create a standard that takes into account modern circumstances by addressing the issue of how creative control is used in AI-mediated processes and go beyond the unquestioned presumption that creativity is intrinsically human. These concerns invite a set of familiar objections, firstly that recognising AI-assisted works would unduly expand copyright and encroach upon the public domain. While sometimes overstated, this concern is not without merit. A control-based standard, however, does not automatically extend protection. Many AI-generated outputs will fall short of the protection level since it requires demonstrable human participation with expression. A second criticism is that rights should be granted to AI developers rather than end users. This claim is difficult to support. Developers design systems, not the individual works those systems generate. To vest authorship in them would detach copyright from the act of creation, transforming it into a form of industrial entitlement. The more serious concern, however, is institutional. A control-based approach runs the danger of generating uncertainty while raising adjudication costs, and courts are ill-suited to evaluate creative processes at the fine level. This concern is real, reflecting a genuine tension between doctrinal accuracy and administrability.
Yet the alternative is no less problematic, retaining a vague and under-theorised requirement of human authorship does not eliminate uncertainty; rather, it is just hidden beneath the façade of clarity. Therefore, rather than choosing between simplicity and precision, the decision is between opposing forms of indeterminacy. A control-based approach has the advantage of, at the very least, making its criteria explicit as opposed to implicit and unanalysed. The reliance on human authorship devolves into assertion in the absence of a control theory. It indicates that people must be important, but it doesn’t explain how or why. The requirement for human authorship is not eliminated by artificial intelligence, but it does compel the legislation to clarify what it means. Unless courts are prepared to articulate a standard grounded in control over expression, decisions such as Thaler v. Perlmutter will continue to resolve peripheral cases while leaving the central problem unaddressed. The future of copyright lies not in preserving the language of human authorship, but in giving that language real substance.
Author: Ms. Anwita Parashar, student at NALSAR
Link to similar articles: https://jpassociates.co.in/ai-generated-content/
1) Thaler v. Perlmutter, 687 F. Supp. 3d 140 (D.D.C. 2023).
2) Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340 (1991).
3) Eastern Book Co. v. D.B. Modak, (2008) 1 S.C.C. 1 (India).
4) Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884).
5) Jane C. Ginsburg & Luke Ali Budiardjo, Authors and Machines, 34 Berkeley Tech. L.J. 343 (2019), https://scholarship.law.columbia.edu/faculty_scholarship/2323.