Who Owns AI-Generated Works? Authorship, Originality and What It Means for Businesses under Indian Copyright Law

Who owns what an AI produces?

Marketing teams are generating campaign images with a prompt. Developers are shipping code written with an assistant. Designers are iterating on AI-produced mood boards. Each of these outputs raises the same question: does anyone own the copyright, and if so, who? The answer under Indian law is neither “nobody” nor “whoever typed the prompt”. It depends on how much human creative input went into the work, and on a statutory provision written long before generative AI existed.

This article explains the current position, including the Copyright Office’s August 2026 DABUS decision and the July 2026 Delhi High Court ruling in ANI v. OpenAI, compares it with the United States, the United Kingdom and China, and sets out practical steps for businesses that use AI output. The position is as of 5 November 2026 and is changing quickly.

What the Copyright Act requires

Copyright in India subsists in original literary, dramatic, musical and artistic works, and in cinematograph films and sound recordings (Section 13). Courts require originality, which the Supreme Court in Eastern Book Company v. D.B. Modak (2007; (2008) 1 SCC 1) described as independent creation with at least a minimal degree of creativity.^{[1, 2]} The Act is built around an author, and the author is generally the first owner (Section 17).^[1]

The one provision that speaks directly to machines is Section 2(d)(vi). For a literary, dramatic, musical or artistic work that is computer-generated, the author is “the person who causes the work to be created”.^{[1, 3]} Three points about it matter. It was inserted in 1994 with computer-assisted tools in mind, not autonomous generative systems, and we are not aware of a court decision applying it to generative AI. It covers only literary, dramatic, musical and artistic works, so films and sound recordings are outside its terms. And it leaves open who “causes” the work when an AI system is involved: the user who writes the prompt, the developer who built the system, or someone else.

Where things stand in 2026

The Copyright Office’s DABUS decision. In 2022 Stephen Thaler applied to register the artwork “A Recent Entrance to Paradise” and named his AI system DABUS as its author. The application was pending for almost four years. On 9 April 2026 the Delhi High Court directed the Registrar of Copyrights to decide it, preferably within eight weeks of a hearing listed for 27 April 2026. It dealt only with the delay and did not rule on the merits.^{[4, 3]} The Registrar’s order is dated 31 August 2026 and was reported in September.^{[3, 5]}

The order made two separate findings. First, it found the work original under Section 13: the fact that it was generated by an algorithmic process did not, by itself, make the expression unoriginal. Second, it held that an AI system has no legal personality, so it cannot be an author under Section 2(d)(vi), and the Legislature did not identify the computer, software or generative system as the author. The Registrar treated Thaler as the person who caused the work to be created, said the defect in the application was curable by naming him as author, and declined to register the work after Thaler declined to amend.^{[3, 5]} We have read the order through these published accounts, not the order itself.

Read together, this is narrower than “AI works cannot be protected”. Authorship sits with a human, and the machine is treated as an instrument.

The unanswered question. The order settles who cannot be named as author. It does not say how much human involvement is enough, and in particular whether typing a prompt into a commercial tool is enough to “cause” the work to be created, which is the question most businesses actually face. It records that the degree and character of human involvement may vary materially from one system and one work to another.^[5]

An inconsistency on the register. In 2020 the Copyright Office registered the artwork Suryast with the AI painting app RAGHAV listed as a co-author, and in November 2021 it issued a notice of withdrawal. Reports differ on whether the registration still stands, and we have not been able to confirm its current status. Anyone relying on registry practice should be cautious.^{[6, 7]}

The government’s work in progress. In December 2025 a DPIIT committee published Part 1 of a working paper on generative AI and copyright. It deals with training data and proposes a mandatory blanket licence with a statutory remuneration right. It set aside the copyrightability and authorship of AI outputs for a Part 2.^[8] We could not find Part 2 at the date of writing, and the DPIIT website should be checked.

ANI v. OpenAI (24 July 2026). In India’s first significant generative AI copyright ruling, the Delhi High Court (Amit Bansal J.) refused ANI an interim injunction against OpenAI. At the prima facie stage it held that storing ANI’s works to train the model fell within fair dealing, found no prima facie evidence of memorisation or substantial reproduction in the examples before it, and left retrieval-augmented outputs undecided because they had not been pleaded. The ruling is interim and the suit continues. It is mainly about training, not ownership of output, so it should not be read as settling who owns what an AI produces.^[9] For a full analysis see our case study on AI and copyright in India.

A spectrum, not a binary

Until the courts or DPIIT say more, the safest way to analyse a work is by the degree of human creative contribution. The assessment below is ours and is not settled law.

AI as a tool: strong. A human drafts, composes or designs and uses AI to assist, edit or refine. Human authorship is clear, and originality is judged on the human contribution.

Substantial human direction: reasonable, but uncertain. Many iterations, curated and selected outputs, then edited, arranged or combined. Human creative choices are present, but whether Section 2(d)(vi) or ordinary authorship applies is untested.

A single short prompt, output used as generated: uncertain. The Registrar has not said whether a prompt “causes” the work, so the question is open.^[5]

Output generated autonomously with no human creative input: weak. No human author can be identified, and an AI system cannot be named as author.^[3]

How other jurisdictions compare

United States. The Copyright Office’s January 2025 report concluded that, given current technology, “prompts alone do not provide sufficient human control to make users of an AI system the authors of the output”. Human contributions can be protected, such as original inputs that are perceptible in the output, creative selection and arrangement, and human modifications.^[10] The D.C. Circuit confirmed in Thaler v. Perlmutter (18 March 2025) that the Copyright Act requires human authorship, and the US Supreme Court denied certiorari on 2 March 2026. That case concerned a work with no human author at all, so it does not say how much human input is enough.^[11]

United Kingdom. Section 9(3) of the Copyright, Designs and Patents Act 1988 deems the author of a computer-generated work to be the person by whom the arrangements necessary for its creation are undertaken. India’s Section 2(d)(vi) resembles it in spirit. In its March 2026 report the UK Government said that most respondents to its consultation supported removing this protection, and proposed that it be removed while copyright continues to protect works created with AI assistance. That is a proposal, not a change in the law.^[12]

China. In Li v. Liu (27 November 2023) the Beijing Internet Court protected an image generated with Stable Diffusion. The user had chosen and ordered more than 150 prompts, set parameters and kept adjusting them until the image matched his conception, and the court held that he was the author, not the AI model or its designers. It is a first-instance decision, and Chinese courts do not follow binding precedent.^[13]

The common thread is the human contribution. India’s statutory text is arguably more accommodating than the US position, but the Registrar’s approach has not been tested by a court, and documenting the human contribution is the most useful thing a business can do.

What this means for businesses

Assume protection may be limited. Do not assume that AI-generated logos, marketing images, copy or code will be protected by copyright, or that you can stop a competitor copying them. Where the work is commercially important, build in demonstrable human authorship. Keep records of prompts, iterations, selection decisions and edits: if ownership is ever contested, evidence of human creative input is the strongest answer.

Read the AI platform terms. A platform’s terms may say the user owns the output. That is a contractual allocation of whatever rights exist, and it cannot create copyright that the law does not recognise. Check restrictions on commercial use, the treatment of confidential inputs and any indemnity for infringement claims, which often matter more in practice than the ownership clause.

Fix your contracts with employees, agencies and freelancers. Make sure IP assignment clauses cover works created with AI assistance, and that assignments are in writing and signed, as Section 19 requires.^[1] Require disclosure of AI use in commissioned work, with warranties of originality and non-infringement. A client who pays for a bespoke logo may not want to learn later that it was machine-generated and weakly protected. Adopt an internal AI-use policy covering approved tools, what data may be entered, and review of outputs before release.

Manage infringement risk on the output side. Owning an output does not make it safe to use. An AI output may reproduce or closely resemble someone else’s protected work, and the ANI ruling did not clear every output claim.^[9] Run reasonable checks on high-value outputs, especially images, music and long-form text.

Use other forms of protection. Rely on trade marks for brand names and logos, confidentiality for sensitive material and contracts for exclusivity. For products, remember that under Section 15 of the Copyright Act copyright in a design ceases once the design is applied to an article reproduced more than fifty times by an industrial process, so registered design protection may matter.^[1]

If you are registering a work. Do not name an AI system as author or co-author. Identify the human author and describe the human contribution accurately, because the Registrar has refused to register an AI system as author.^{[3, 5]}

Our intellectual property practice advises on authorship, licensing and AI-related contracts; see also our article on IP enforcement in the generative AI era. This article is general information, not legal advice.

Frequently asked questions

Can an AI system be the author of a work in India? No. In the DABUS decision the Registrar held that an AI system has no legal personality and cannot be named as author under Section 2(d)(vi).

Who owns an image generated from my prompt? The law has not settled this. Section 2(d)(vi) makes the author of a computer-generated work “the person who causes the work to be created”, but the Registrar has not said whether a prompt is enough. The more you select, edit, arrange and direct, the stronger the claim.

Does the platform’s terms of use give me copyright? No. The terms can allocate whatever rights exist between you and the platform, but they cannot create copyright that the Copyright Act does not recognise.

Does ANI v. OpenAI decide who owns AI output? No. The July 2026 interim ruling concerns the use of copyrighted works to train an AI model, not ownership of what the model produces.

What should I record to protect an AI-assisted work? Keep your prompts, iterations, choices between outputs and the edits you made, since evidence of human creative input is the strongest answer if ownership is contested.

Outlook

Three developments are worth watching: the publication of DPIIT’s Part 2 on AI outputs, any proposal to amend Section 2(d)(vi), and the first court ruling on whether prompting alone makes a person the author. Until then, the sensible approach is to treat AI as a powerful tool, keep a human meaningfully in the creative loop, and let contracts do the work the statute does not yet do.