The Battle Over AI Authorship
The Illusion of Creation
You’ve 45 minutes to create the ideal prompt, a particular emotional state or heat, References to Vermeer’s light and Brutalism’s geometry. You press enter and either Midjourney or Google’s Gemeni gives you something that’s truly arresting, that feels like yours in your bones. It’s a true phenomenon. Legal safeguards it implies? Almost certainly not. The biggest psychological pitfall of the age of generative AI is the feeling of authorship is very real, and the legal system of authorship is just as stubborn as it is indifferent. The hours of iteration, refinement, rejection and redirection in copyright terms might be equivalent to what you do in a very sophisticated search engine. It is here that the commission analogy is helpful. When you hire a painter, and inform him what is on the canvas, what colour palette you want, the style of painting, and the mood, the painting is still the painter’s painting. You might be the rightful proprietor of the body of the canvas. It’s possible that you have usage rights. But the authorship and copyright is attached to the human hand that held the brush. Then use a diffusion model instead of the painter. The instruction set gets more complex, but the basic question turns into a really awkward version: When is it that someone is directing a machine rather than creating it?
Worldwide, the answer to copyright is rather blunt. You will most likely not enjoy it.
Exclusive Club of Human Authorship (ECHA)
Copyright was never meant to be “liberal Copyright was never meant to be generous”. It serves as a nudge to promote human creativity, by allowing creators to monopolize expression for a period of time. Always has been humankind. This is a position that has become entrenched, both in policy and by litigation, in the United States. The U.S. Copyright Office has considered, and denied, numerous applications for copyright registration for AI-generated works that did not meet the standard for “substantial human involvement”. The U.S. Copyright Office has been rejecting applications to register AI-generated works on several occasions for lack of “substantial human involvement”. In the precedent-setting Thaler v. Perlmutter case (D.D.C., 2023), it was clear that an AI system cannot be the author and that a work created autonomously by an AI is not eligible for copyright protection. The court found its reasoning in centuries-old doctrine that copyright begins with human creativity. Full stop. As was characteristic, the United Kingdom made a more pragmatic exception. Under section 9(3) of the Copyright, Designs and Patents Act 1988 (CDPA), each person who carries out the “arrangements necessary for the creation of the work” is the copyright owner of a computer-generated work. No human author needed a human arranger. This so called “Maverick Exception” is one of the few jurisdictions that expressly considers machine-generated works as protected. The UK is one of the few jurisdictions that explicitly contemplates machine-generated works as protected – this is the “Maverick Exception”. In the AI era, its existence is important, and it is being challenged; the term of protection is lowered (50 years vs. 70 years +); there are doubts about its existence. It is a clear statement in the law that the human-machine boundary would always be a fluid one. The majority of the world operates somewhere in the middle: they take the analog concepts to the digital world and expect things to work out without the analog/digital disconnect becoming too apparent in the courtroom. That tension is extremely combustible at the moment in India.
India’s 2026 Reckoning
The Indian Copyright Act, 1957 has an interesting clause which could not have been expected to be applicable to generative AI. In the case of a computer-generated work, the definition of “author” is as the person who causes the work to be created under Section 2(d)(vi) of the Act. This is a no-brainer, as it appears to be a win for the AI prompt engineers. However, it requires a lot of strain to handle the modern autonomous generation. “Causation” of a work suggests some sort of causal agency with meaning (intent, direction, selection). With a large language model built on terabytes of human expression, synthesizing an image, or a symphony, from a statistical prompt is similarly diffuse, probabilistic, and deeply contested. Does the click of “generate” mean that the person actually caused it? It is about to be revealed to Indian courts. The first time the judiciary was seen to be uncomfortable was in the RAGHAV AI co-authorship case. This was a truly radical approach by the Copyright Office, which had initially registered a work listing RAGHAV, an AI tool, in conjunction with a human creator, but later retracted. This showed an institution improvising on-the-fly, without any statutory roadmap to follow. The more significant battle took place in April 2026 when the Delhi High Court directed the Copyright Office to finally determine the fate of Stephen Thaler’s long delayed application to register “A Recent Entrance to Paradise” the visual artwork that was created wholly and exclusively by Thaler’s DABUS AI system, without any human creativity whatsoever. In all proceedings before the United States, the United Kingdom, Australia and India, Thaler has contended that he, “the person who caused the work to be created,” under Section 2(d)(vi) is the relevant person. The court’s order, which is a de facto order to the Copyright Office, that it should move off the fence is setting up the decision which could overturn the regulatory regime in India’s entire AI-IP landscape. An appeal will likely follow and the ultimate reasoning of the judiciary will be the bedrock on which the Copyright Office’s reasoning will be based. A parallel, and perhaps more commercially important battle is raging, meanwhile. The Delhi High Court is hearing the case of ANI Media vs. OpenAI, which challenges the input component of the AI pipeline in particular: the news content used to train big models is copyrighted, and therefore, the fact that it is used is widely accepted as theft, which is a “form of extractive creativity” that is “laundering” the protected expression into statistical weights. Ownership of output and liability for inputs are separate legal issues which are all aspects of the same disruption of the ecosystem. At the same time, India is facing questions from those who want to know who benefits from these systems, as well as who loses their rights as they’re created.
The Continuum of Human Involvement
AI-generated creation isn’t created equal. Courts around the world are now starting to draw a sort of continuum and your work will fall somewhere along those lines and that’s an enormous difference.
| Level | Human Involvement | Likely Protectability |
| Pure Generation | Single prompt, no iteration, no editing | Unprotectable public domain |
| Iterative Prompting | Multiple rounds, deliberate rejection and selection | Marginal highly jurisdiction-dependent |
| Curated Compilation | Selecting, arranging, and sequencing AI outputs | Compilation copyright possible (thin protection) |
| Post-Processing | Substantial manual editing, retouching, or remixing of AI output | Stronger claim derivative work potential |
| AI-Assisted Original | AI used as a tool within a predominantly human creative process | Protectable AI analogous to Photoshop or a camera |
What’s important is the documented demonstrable human creative judgment. A creator who can provide a prompt history, with their iterative process, a time-stamped sequence of their own creations, the deliberate choices they made in the process, and what they rejected and what they went back to and maybe some of their own personal editing and then some additional refinement and some new iteration will be in a categorically different legal position than one who can’t. This is backed by the U.S. Copyright Office’s own rules and guidelines, which determine the copyrightability of works created using AI by examining whether the output has been sufficiently “authored by the human author in the selection, arrangement and creative shaping of the work. The threshold is not very high per se. However, it needs to be substantiated.
The Commercial Warning Shot
What most startups and commercial creators are not factoring in is this: If an AI output isn’t copyrightable, then it is in the public domain, as soon as it is created. That means your competitor can legally use your AI-generated brand imagery, product images and marketing copy the very day you publish it without paying you anything, without giving you credit. Just because you’re not infringing doesn’t mean that you can’t claim copyright. This is a vulnerability for companies seeking to create commercial value from AI-generated creatives. But the law is not moving towards resolution fast enough to be comfortable. The need for action is thus pressing and specific: Documentations are needed. Keep notes on your rounds and rounds of prompts. Note down the decisions of rejection and selection. Export post processing history. Enhance AI-generated content with additional originality, particularly for smaller elements like headlines and subtitles. Develop a paper trail and show, at every step, the use of human skill and judgment. The modicum of creativity standard was retained from the Feist Publications v. Rural Telephone Service (US, 1991) case, which has a low threshold. Courts don’t want to see genius. They are seeking evidence of a human subject, whose true creative decisions, created a final expression. That proof can be created only with difficult effort and great legal peril after the fact. The days of “assuming” creative works are over. With generative AI innovations, authorship needs to be crafted just as carefully as the product is created.
References
Thaler v. Perlmutter — U.S. Human Authorship Requirement
UK Section 9(3) CDPA — The “Maverick Exception”
Indian Copyright Act, Section 2(d)(vi)
The RAGHAV AI Co-Authorship Flip-Flop
ANI Media v. OpenAI
