AI-Generated Content and Copyright in India: Who Owns the Rights?

INTRODUCTION

When an artist feeds a photograph and a style reference into an AI painting tool and receives back a finished canvas, or a user types three lines into an image generator and receives a photorealistic result, Indian copyright law is asked a question it was not built to answer with any real precision: who, if anyone, owns what came out? The Copyright Act, 1957 was drafted around a human creator exercising skill and judgment. Generative AI systems — large language models, diffusion-based image generators, AI music composers — now routinely produce output with only a prompt, a few parameters, or a curated dataset standing between the human and the finished work.

India has not enacted, and has shown no present intention of enacting, an AI-specific authorship or ownership regime. Unlike the position on training-data infringement — addressed, at least provisionally, by the Delhi High Court's interim order in ANI Media Pvt Ltd v OpenAI OpCo LLC on the fair-dealing question¹ — the ownership question is being worked out through a single, thirty-year-old statutory provision and a handful of administrative and judicial proceedings that remain, in the most literal sense, unfinished business. This article examines that provision, the two disputes that have tested it, and the practical layer of contracts and platform terms that sits above the unresolved doctrine.

THE STATUTORY ANCHOR: "THE PERSON WHO CAUSES THE WORK TO BE CREATED"

Section 2(d) of the Copyright Act, 1957 defines "author" differently for each category of work — the writer for a literary work, the composer for a musical work, the artist for an artistic work.² Sub-clause (vi), inserted by the Copyright (Amendment) Act, 1994,³ was drafted for a narrower and more mundane purpose than generative AI: to fix authorship in works produced by early computational tools, spreadsheet outputs, word-processed documents and algorithmically arranged databases. It provides that for a literary, dramatic, musical or artistic work "which is computer-generated," the author is "the person who causes the work to be created."

On its face, the provision appears to resolve the AI-authorship question by locating a human — the prompter, the developer, the person who configures the system — as author. In practice, the difficulty is that the 1994 Parliament could not have contemplated a system that generates a photorealistic image from a three-word prompt with no meaningful human creative contribution to the final expression. "Causing" a deterministic spreadsheet macro to run and "causing" a diffusion model to synthesise an image are not obviously the same act, and the Copyright Office has issued no rulemaking or policy statement clarifying where the line sits.⁴ Originality itself is a separate threshold under Section 13, and the applicable judicial standard — a minimal degree of independently generated, non-trivial creative expression, short of the higher "sweat of the brow" threshold rejected in Eastern Book Company v D.B. Modak⁵ — is a distinct inquiry from who counts as author. As the two disputes below show, a work can clear the originality bar while its authorship, and therefore its ownership, remains genuinely contested.

ORIGINALITY WITHOUT AUTHORSHIP: THE REGISTRAR'S REJECTION OF DABUS

India's first reasoned administrative ruling squarely on generative-AI authorship came on 31 August 2026, when the Registrar of Copyrights, Dr Unnat P Pandit, rejected an application by US computer scientist Dr Stephen L Thaler to register the artwork "A Recent Entrance to Paradise" with his AI system, DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), entered as author.⁶ The application, filed on 3 May 2022, had named DABUS as author and Thaler as the person claiming ownership. A discrepancy letter issued on 31 July 2023 asked the applicant to identify a legally competent author; unsatisfied with the response, the Copyright Office convened hearings on 29 April, 25 May and 24 June 2026, appointing Senior Advocate Ms Rajeshwari Hariharan as amicus curiae on 13 May 2026 — a course prompted in part by the Delhi High Court's order of 9 April 2026 directing the Registrar to decide the long-pending application within eight weeks.⁷

The Registrar's reasoning proceeded in two distinct steps that matter for the ownership question. First, on originality, the Office accepted the amicus curiae's submission that the Section 13 inquiry — as explained in Eastern Book Company v D.B. Modak — requires only a minimal degree of independently generated, non-trivial creative expression, and that the work's arrangement of visual and compositional elements met that threshold.⁸ Autonomous or AI-assisted generation, in other words, does not by itself disqualify a work from copyright protection.

Second, and separately, on authorship, the Registrar held that DABUS "is not a natural or juristic person recognised in law and cannot be entered as author under Section 2(d)(vi)." The Office expressly rejected the analogy Thaler's counsel drew to companies or Hindu idols, which Indian law does recognise as juristic persons: DABUS possessed no statutory recognition, proprietary status, civil personality or legal capacity independent of its creator.⁹ Turning to who, then, had "caused" the work to be created, the Registrar rejected the applicant's argument that DABUS, as the "operative cause" performing the immediate generative act, displaced Thaler as merely the "upstream cause." Drawing on the "mastermind" or effective-cause test associated with Burrow-Giles Lithographic Co v Sarony and Aalmuhammed v Lee, the Registrar found that Thaler — who conceived and configured DABUS, supplied the visual inputs and curated the linguistic parameters — was himself the person who caused the work to be created, and had been offered, and had declined, the opportunity to amend the application to name himself as author.¹⁰ "Autonomy in execution," the order observed, "is not synonymous with conception of a work"; copyright law "recognizes the mastermind behind the creation of the work, not the autonomy of the tool through which that creation is implemented."¹¹

The order is narrow by design.¹² It does not decide whether AI-generated works are copyrightable as a class, and it turns on facts — a single, heavily human-directed generative process — that will not map neatly onto every AI use case, particularly large language model or diffusion outputs generated from a short prompt with comparatively little iterative human curation. What it does settle, provisionally, is that an AI system itself can never be the "person who causes the work to be created" under Indian law, and that a human standing sufficiently close to the generative process — as conceiver, configurer and curator — ordinarily will be.

THE SURYAST PRECEDENT: ANKIT SAHNI, RAGHAV, AND THE UNRESOLVED QUESTION OF JOINT AUTHORSHIP

The harder ownership question — not whether an AI system can be sole author, but whether it can be a co-author alongside the human who directed it — was raised years earlier and remains formally unresolved. In 2020, lawyer and artist Ankit Sahni supplied his own photograph and Vincent van Gogh's The Starry Night as a style reference to RAGHAV (Robust Artificially Intelligent Graphics and Art Visualizer), an AI painting application, producing a stylised work titled Suryast. Sahni applied to register the work naming himself and RAGHAV as co-authors, on the basis that RAGHAV's contribution — its choice of colours, brushstrokes and stylistic transformation — was "distinct, disparate and independent" from his own.

In November 2020, the Copyright Office granted the registration as filed — India's first, and to date only, copyright certificate naming an AI system as a co-author.¹³ The registration attracted international attention precisely because it appeared to break from the human-authorship orthodoxy. The Office subsequently reversed course, issuing a withdrawal notice that invoked Sections 2(d)(iii) and 2(d)(vi) of the Copyright Act and reasoned that an author must be a natural person capable of holding and exercising rights — a status RAGHAV, an AI application, could not have. Sahni contested the withdrawal, arguing the Copyright Office lacked jurisdiction to reconsider its own completed registration. That dispute is presently pending before the Delhi High Court and remains undecided.¹⁴

The comparative record only sharpens the stakes of that pending decision. The United States Copyright Review Board rejected Suryast's registration in December 2023, holding it "lacks the human authorship necessary to support a copyright claim." Canada's Intellectual Property Office took the opposite view, issuing registration number 1188619 in December 2021 listing Sahni and RAGHAV as co-authors — a registration itself now under legal challenge in the Canadian Federal Court.¹⁵ India, having first granted, then withdrawn, and now left the question to its High Court, occupies an unusually exposed middle position: unlike the DABUS order, which foreclosed AI sole authorship relatively cleanly, the Suryast litigation goes to whether an AI tool's contribution can ever be legally significant enough to generate joint ownership — a question the DABUS Registrar's order does not purport to answer, since Thaler never pleaded joint authorship.

OWNERSHIP BEYOND AUTHORSHIP: FIRST OWNERSHIP, CONTRACTS, AND PLATFORM TERMS

Authorship and ownership are distinct inquiries under the Act, and the practical allocation of rights in AI-generated content in India today rests as much on Section 17 and private contract as on Section 2(d)(vi). Section 17 vests first ownership in the author save for specified exceptions — commissioned works, and works created by an employee in the course of employment, where the employer is ordinarily the first owner absent a contrary agreement. Because an AI system cannot be an author, it equally cannot be an "employee" for these purposes, so the employer-employee and commissioning exceptions that would normally settle ownership disputes over human-assisted work offer no direct analogy for AI output; the DABUS order confirms as much in holding that a non-juristic system can neither be first owner under Section 17 nor validly assign rights under Sections 18 and 19.¹⁶

In the absence of statutory allocation, ownership of commercially significant AI-generated content in India is, for now, largely a matter of contract layered on top of an uncertain authorship baseline. Employers routinely address AI-assisted work product through employment and work-for-hire clauses; commercial users of AI platforms are bound by the platform's own terms of service, which vary considerably — some providers purport to assign all rights in generated output to the user, others reserve residual rights or impose use restrictions, and few are drafted with Indian authorship doctrine specifically in mind. A business relying solely on a platform's terms of service to establish ownership, without a human contributor who can plausibly satisfy the Section 2(d)(vi) "causes to be created" test, is relying on a contractual promise that Indian courts have not yet been asked to test against the statute. Until either the Suryast litigation or subsequent Copyright Office practice clarifies where the line of sufficient human contribution sits, the safer registration and commercialisation practice — consistent with the practical guidance now emerging from Indian IP practitioners — is to identify a clearly established human author, evidencing the specific choices, curation and configuration that satisfy the causation test the DABUS order applied, rather than naming an AI tool as author or co-author.

CRITICAL ANALYSIS: THREE UNRESOLVED QUESTIONS

Three gaps deserve particular scrutiny. First, the DABUS order's "mastermind" test was applied to an unusually human-directed generative process — curated inputs, configured parameters, an inventor-operator relationship the Registrar could readily trace. Large language model and diffusion-model outputs generated from a short natural-language prompt, with the model exercising substantial unsupervised choice over composition, style and detail, sit less comfortably within that same reasoning; the order itself acknowledges it decides only the application before it, leaving open how the causation test applies where human direction is thin.¹⁷

Second, the Suryast litigation raises a question the DABUS order does not reach at all: whether Indian law can ever recognise an AI system's contribution as sufficient to generate joint authorship, short of sole authorship. A ruling either way will have consequences well beyond visual art — AI-assisted music, legal drafting, software and journalism all raise the same structural question of whether a human collaborator's contribution can be legally "diluted" by a non-human co-contributor without extinguishing protection altogether.

Third, the practical reliance on platform terms of service to allocate ownership sits on an unexamined assumption: that a contractual assignment of rights in output is effective even where the putative assignor — the AI platform — could not itself have held authorship or ownership in the first place under Section 2(d)(vi). If an AI system cannot be an author, and a platform provider is not, on ordinary facts, the person who "causes" a particular user's output "to be created," the doctrinal basis on which a platform's terms of service purport to transfer rights it may never have held remains untested in Indian courts.

THE WAY FORWARD

Three steps would meaningfully reduce the present uncertainty without requiring wholesale legislative reform. First, the Delhi High Court's eventual ruling in the Sahni matter should, ideally, articulate a workable standard for how much human curation, selection or configuration is sufficient to satisfy Section 2(d)(vi) in an AI-assisted, as opposed to AI-directed, context — building on, rather than departing from, the causation reasoning the Registrar applied to DABUS. Second, the Copyright Office would benefit from issuing practice guidance, akin to the approach taken by the US Copyright Office in its own AI policy statements, clarifying what evidence of human authorship it expects in registration applications involving generative AI tools, reducing the case-by-case discrepancy-letter process that took the DABUS application more than four years to resolve. Third, given how much of the practical ownership landscape currently rests on platform terms of service, MeitY and the Copyright Office might usefully examine, as part of the wider AI Governance Guidelines process, whether such contractual assignments require any statutory backstop where the assignor's own underlying rights are themselves uncertain.

CONCLUSION

India's answer to "who owns AI-generated content" currently rests on a single, three-decade-old provision drafted for spreadsheets and databases, one reasoned administrative order that firmly excludes the AI system itself from authorship while leaving open how far the human "mastermind" test stretches, and one still-pending High Court matter that will determine whether joint authorship with an AI tool is possible at all. The DABUS order supplies welcome clarity on one narrow point — originality can survive heavy AI involvement, but authorship cannot vest in the machine — while leaving the Suryast question, and with it the harder commercial reality of AI-assisted creative and business output across India, genuinely open. Until the Delhi High Court rules, or Parliament revisits Section 2(d)(vi) with generative AI specifically in view, ownership of AI-generated content in India will continue to be negotiated, case by case, at the uneasy intersection of a 1994 statute, a 2026 administrative order, and the fine print of AI platforms' terms of service.

REFERENCES

Legislation

Copyright Act 1957.

Copyright (Amendment) Act 1994.

Cases and Administrative Orders

Aalmuhammed v Lee, 202 F3d 1227 (9th Cir 2000).

ANI Media Pvt Ltd v OpenAI OpCo LLC, Neutral Citation 2026:DHC:5900 (Delhi High Court, 24 July 2026).

Ankit Sahni v Registrar of Copyrights (Suryast registration withdrawal), pending before the Delhi High Court.

Burrow-Giles Lithographic Co v Sarony, 111 US 53 (1884).

Canadian Intellectual Property Office, registration no. 1188619 (1 December 2021); challenge pending, Federal Court of Canada (CIPPIC v Sahni).

Eastern Book Company v D.B. Modak (2008) 1 SCC 1.

In the matter of Diary No. 9356/2022-CO/A, Applicant: Dr Stephen L Thaler, before the Registrar of Copyrights (order dated 31 August 2026).

United States Copyright Review Board, decision on 'Suryast' (11 December 2023).

Secondary Sources

Ahlawat & Associates, 'AI Generated Content Copyright in India' (2026).

Asia IP, 'Delhi High Court to copyright office: Decide on Thaler application for AI artwork in eight weeks' (2026).

Bar and Bench, 'AI-generated work can get copyright but AI system cannot be author: Indian copyright office' (2026).

Business Standard, 'Copyright Office rejects AI system DABUS as author under Indian law' (2026).

Chambers and Partners, 'Delhi High Court's ANI v OpenAI: Shaping AI Copyright Law in India' (2026).

CIPPIC, 'CIPPIC v Sahni' project page.

Holland & Knight, 'Generating a Body of Generative AI Case Law' (2023).

Footnotes

Khurana & Khurana, 'AI-Generated Content and the Copyright Act 1957: Is There a Gap in Indian Law' (2026).

Prime Legal, 'AI Copyright Law India: The Complete 2026 Guide' (2026).

SpicyIP, 'Originality Without Authorship: Analysing the Copyright Office's Decision in Thaler' (2026).

Storyboard18, 'India says AI cannot author a work, but it can still protect one' (2026).