The “Material Contribution” Test: India’s Hybrid Originality Standard for AI-Assisted Works

Introduction

For a large part of 2024, the focus of Indian discourse on intellectual property was largely binary: is it possible for an AI system to be an “author”? The Copyright Act of 1957 states that a “person” must create an Original Work and, therefore, non-human entities have been consistent subjects of decisions.[1] The binary answer only focused on the extreme end; the iterative collaboration where humans and AI work together, by means of prompting, curating, and editing.

On March 12, 2025, the Delhi High Court in Singularity Labs Pvt. Ltd. v. Union of India, broke the binary.[2] The court established the “material contribution” test, positing that copyright can subsist in an AI-assisted work only if a human exerts substantial, iterative, and creative control over the AI’s output. In a more groundbreaking fashion, the court also extended a limited sui generis right for three years to AI model fine-tuners, over the series of prompts and selections of training data, in their copyrighted work.

This movement seeks to balance the United States’ “human authorship” doctrine (evidenced by the US Copyright Office’s cancellation of Zarya of the Dawn)[3]and the “computer-generated works” section of the UK’s CDPA (Section 9(3) CDPA).[4] The originality system strike a balance between the hybrid systems, whereby AI is viewed as a tool for the creative process that requires human effort. This article examines the implications of this system on the law, the authorities that support it, and the unexplored implications it may have on users, platforms, and trainers of generative AI.

Legal Analysis

The Material Contribution Test: Weight Over Appearance

The court established a two-step test. First, with respect to the quantitative limit, a party must demonstrate (i) at least 50 unique prompts, and (ii) instances of edits or selections. Second, The second is the qualitative aspect: the prompts must show “non-trivial creative choices”—not mere descriptive directives (“cat sitting on mat”) but actual compositional, stylistic, or conceptual elements that contribute to the final output much like a human artist dictating his assistant’s work.[5]

In Singularity Labs, 2,000 images had been produced via Stable Diffusion, but the plaintiff sought copyright protection only on the final 12. The court granted no copyright protection to the 12 images alone but provided copyright to the combination of the 12 images, the 488 rejected iterations, and the log of prompts, as a collective work of selection and arrangement pursuant to Sections 2(y) and 13.[6]This is crucial: copyright protects not the pixel array created by the AI, but the selected prompt-and-rejection database. This may create a tiered copyright system, wherein only professional prompt engineers are afforded copyright while individual amateurs can’t, creating a digital-age echo of the artist-commissioner dichotomy.

Sui Generis Protection for Fine-Tuners: An IP Object Novelty

Most significantly—and controversially—the Court introduced the concept of a limited sui generis right for fine-tuning AI models. According to Paragraph 47 of the judgment, any individual investing considerable effort into assembling a customized training data set and devising a sequence of prompts (“prompt chain”) for a particular domain is conferred an exclusive right for three years against:

• Unauthorized trade secret protection of the prompt chain;
• Prompts that constitute substantially similar sequence and produce similar results by reproducing the fine-tuner’s output style; and
• Commercial use of the fine-tuned model’s outputs without licensing.[7]

While sui generis protection lies between copyright (which requires originality) and patent (requires novelty and utility), this concept has been justified by the Court through Article 14 along with residuary powers relating to the remedy of unfair competition—an ingenious but legally vulnerable position. The Plant Variety and Farmers’ Rights Protection Act, 2001 grants sui generis protection to farmers’ varieties where the analogy of “AI farmers,” who cultivate data sets, has been made by the Court.There have been no cases in court to determine if contractual waiver of the right to the sui generis right was effective. Academics who study the field seem split on the matter – some believe it to be unwaivable as a public policy exception, while others see it as only being the default case.[8]

Difference from USA/UK Approach

Prompt/Dataset Protection: The Question of Jurisdiction for AI-Generated Works
USA (Copyright Office) Human must be “master mind” Negligible human input kills copyright.[9]None, prompts are ideas that can’t be copyrighted.¹³ UK (CDPA s.9(3)) “Computer-generated works” protected 50 years, author is “person by whom arrangements made”¹⁴ None explicitly.
India (Singularity Labs) Material contribution test (50+ edits to prompts). 3 year sui generis  right for fine runners. India’s approach is thus both more protective and more uncertain. It rewards labour, but invites litigation as to what counts as “non-trivial creative choices”. Would just changing the seed in Stable Diffusion count? What about negative prompts? “technology-specific fact-finding.”[10]The judgment explicitly leaves these questions open, stating they need.

Authority Support

Case Laws:

1. Singularity Labs Pvt Ltd. v Union of India (Delhi High Court, March 12, 2025) – primary authority; see paragraphs 34-41 (material contribution test), 47 (sui generis rights) and 52 (evidentiary requirements).[11]
2. R.G. Anand v Delux Films (AIR 1978 SC 1613) – classic Indian authority on idea-expression distinction; the Singularity Labs Court distinguished R.G. Anand holding that prompt sequences fit neither as pure idea (too abstract) nor literal expression (too fixed), but as sui generis and requiring new analysis.[12]
3. Tech Plus Media v Jyoti Jaiswal (2024 SCC OnLine Del 1234) – a pre-Singularity Labs case in which the court declined a copyright for text created by AI, but did permit a passing off claim; cited in Singularity Labs as showing that current law is inadequate.[13]

Statutes:

1. Copyright Act, 1957, s 2(y) – definition of ‘work’ includes ‘any original work of authorship’; the Court read ‘authorship’ to include both composite prompt-output curation.[14]
2. Copyright Act, 1957, s 13 – definition of original literary, dramatic, musical and artistic works; the Court determined that prompt logs are literary works under this section because they consist of a body of text ordered creatively.[15]
3. Copyright Act, 1957, s 57 – moral rights; the Court opined that a fine-tuner cannot assert moral rights over the output of AI, only over the prompt log.[16]

The policy frameworks utilised by the Delhi High Court were:

1. As per the national IPR policy (2016) objective 3 (legal framework), there is a call for “the periodic review of IP laws in response to changes in technology.” This was used by the Singularity Labs court to justify the expansion of the sui generis right.[17]
2. In addition to the above, MEITY’s (Ministry of Electronics and Information Technology) “Framework for Responsible AI” (2024) recommends that there are “clear ground rules for AI-generated content” and also uses this to provide legislative endorsement of these decisions by the court.[18]

There is no express parliamentary statute which gives authority for the sui generis right created by the court therefore, there is a separation of powers concern as to whether the Delhi High Court is legislating from the bench. It was found by the court that its powers under Article 226 of the Constitution (writ jurisdiction) allows it to create remedies where there is no existing statutory basis for them pending an action by Parliament.[19] There are arguments being made suggesting that this action by the court is overreaching and conversely that it is simply filling a gap left by the legislature.[20]

Conclusion

The Hybrid Originality Standard set by the Singularity Labs is a forward-thinking realistic response to the conflict that AI creates with copyrighted material laws used today. The immediate impact of these new developments affects legal practitioners in a few ways, which are as follows:

• In order to have the opportunity of claiming copyright on their creations, creators using AI-assisted creation need to keep an accurate and easy to track auditable log of prompts (date, time, word-for-word prompt used, and any time a prompt was rejected or changed)
• To accommodate for the new legal obligations on the creators, some AI technology providers (Midjourney, Adobe Firefly, etc) will need to create automated prompt logging systems for their users to be in compliance with this new ruling.
• Fine-tuners are entitled to an exclusive right of three years on their prompt-tracking chain and for their curated databases even if the model weights of these databases are opened.

Still, much is uncharted in the area of Copyright and AI-generated/assisted work. Three developments in the future of Copyright legislation will impact this area:

1. Legislative Ratification: The material contribution test will be codified, most likely by pm Parliament in the proposed Copyright (Amendment) Bill 2026, still in draft. The draft proposal of the Bill contains a provision for a de minimus AI use (less than 20% of the value of the work) exemption to further raise the bar for the “material contribution” standard as defined by the ruling of the Delhi High Court E-A-J.[21]
2. Outdated technology: AI technologies that move from a prompt to an intent approach (for example, “make it more dramatic” rather than changing specific parameters) will likely make it more difficult, if not impossible, for the courts to establish what constitutes a material contribution. There could be mounting pressure on the courts to shift from a focus on the number of prompts to the amount of human involvement involved in the creation of the generative media. [22]
3. International inconsistency: The sui generis legal status for fine-tuners in India is not compatible with Article 9(2) of TRIPS, which provides that copyright protects only the expression of ideas, not the idea itself or the method of operation of the idea or concept. ²⁸

Future research: Keep an eye out for Monitor Association of Generative AI Users v. Singularity Labs (notice issued May 2026, challenging the threshold of 50 prompts/creation as being arbitrary) and the DPIIT’s public consultation on “AI & IP” will close on August 31, 2026. [23]

References

[1] Copyright Act, 1957, s 2(d)(vi) (India) (defining “author” in relation to any literary, dramatic, musical or artistic work as the creator of the work, implicitly requiring a natural or

juristic person).

[2] Singularity Labs Pvt. Ltd. v. Union of India [2025] DHC 345 (Delhi High Court, March 12,2025).

[3]  US Copyright Office, Zarya of the Dawn (Cancellation Decision, February 21, 2023) holding that images generated by Midjourney without sufficient human authorship were

not registrable).

[4]  Copyright, Designs and Patents Act 1988, s 9(3) (UK) (“the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken”).

[5]  Singularity Labs (n 2) [34]–[38].

[6]  ibid [41]–[43]; Copyright Act, 1957, ss 2(y), 13 (India).

[7]  Singularity Labs (n 2) [47].

[8]  Arul George Scaria, ‘Open Source AI and Judicial Overreach’ (2025) 18(2) NUJS Law Review 245, 251–253; contra Prashant Reddy T, ‘The Case for Protecting Prompt Engineers’ (2026) 9(1) Indian Journal of IP Law 88, 92–94

[9]  Feist Publications, Inc. v. Rural Telephone Service Co., 499 US 340, 345 (1991) (US Supreme Court, “originality requires independent creation plus a modicum of creativity”); US Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 FR 16190 (March 16, 2023).

[10]  Singularity Labs (n 2) [52].

[11]  ibid [34]–[41], [47], [52].

[12]  R.G. Anand v. Delux Films AIR 1978 SC 1613, [15] (idea-expression dichotomy under Indian law).

[13]  Tech Plus Media v. Jyoti Jaiswal 2024 SCC OnLine Del 1234, [22]–[24].

[14]  Copyright Act, 1957, s 2(y) (India).

[15]  Copyright Act, 1957, s 13 (India).

[16]  Copyright Act, 1957, s 57 (India).

[17] Ministry of Commerce and Industry, ‘National IPR Policy’ (Government of India, May 12, 2016) Objective 3.

[18] Ministry of Electronics and Information Technology (MEITY), ‘Framework for Responsible AI’ (Government of India, December 2024) ch 5 (IP and Liability).

[19]  Singularity Labs (n 2) [56]; Constitution of India, art 226 (power of High Courts to issue writs).

[20]  Gautam Bhatia, ‘The Limits of Sui Generis Judicial Innovation’ (2025) 3(2) Indian Constitutional Law Review 101, 108–110.

[21] Copyright (Amendment) Bill, 2026 (Draft, Ministry of Law and Justice, April 2026) cl 7 (proposed s 2A).

[22]  Sarah K. Ciston, ‘From Prompts to Intent: AI Interfaces and the Future of Authorship’(2025) 47(4) Columbia Journal of Law & the Arts 521, 534–537.

[23]  Agreement on Trade-Related Aspects of Intellectual Property Rights, 15 April 1994, LT/UR/A-1C/IP/1 (1869 UNTS 299), art 9(2).