When Law Meets the Machine
You may hear the same silent anxiety when walking into any newsroom or music studio today. AI tools can now reproduce, study, and copy content that previously took weeks or months to create in a few seconds. It feels impressive. It is also uncomfortable. Rethinking DPIIT’s AI copyright model in India
This tension is no longer abstract. This has become a live legal test due to the Delhi High Court case between ANI Media and OpenAI, turning it into a policy question. The Delhi High Court has withheld its decision in the ANI v OpenAI case.[1] And that one case has brought a far greater problem into the limelight. Is the copyright law in India prepared to take this change?
The Department of Promotion of Industry and Internal Trade intervened on 9 December 2025, with a working paper.[2] Its concept is easy to understand. One national licence. A single payment system. AI companies get access to copyrighted material for training. Producers receive compensation in a centralized system.
It sounds neat. Maybe even fair. But is it? It is not only a legal question. It is also concerning the equilibrium. Who gets control? And who gives it up?
The Legal Gap India Cannot Afford
The Copyright Act, 1957, was authored in a very different world. It is explicit on authors, rights, and ownership. But it has one fundamental assumption. That is a human creation.[3]
Section 14 provides authors the right to reproduce.[4] Section 51 states that unauthorised copying is infringement.[5] Section 52 permits fair use, which is limited, primarily in research, review, or criticism.[6] These are confined areas. Large-scale AI training has not been extended by courts.
And now consider the mechanism of AI. It does not read one article. It consumes millions. It is not a copy in the ordinary sense. However, it still requires access, and frequently, that can require temporary or stored copies. And where does that fit?
The Supreme Court in Eastern Book Company v. D.B. Modak made it clear that copyright needs some level of human creativity.[7] This line gains significance when machines are introduced. Simultaneously, creators can hardly prevent the use of their work in this process. Thus, both are at cross purposes. Developers are at risk of the law. Creators face loss of control. And in between stands a law that was never made for this scale or pace.
DPIIT’s Hybrid Model
The DPIIT paper tries to break this deadlock with what it calls a hybrid system.[8]
Developers of AI would receive a blanket licence. This licence permits them to train on any copyrighted material that is legally available to them. One does not have to ask every creator. Creators also have no choice but to opt out.
Developers, in turn, have to pay royalties. Such payments are connected with the commercialisation of AI. The rates would be determined by a government-appointed body. These rates may be reconsidered and even disputed in court. The funds would not be company-to-creator. Rather, it would go through copyright societies and a new central body. Only registered works would be eligible for distribution.
So yes, there is logic here. But there is something wrong as well. The system largely treats access to creative work as a default right of AI companies. The primary compensation turns into payment. Control is pushed into the background. And that brings a quiet, solemn question. Is it enough money when the ownership is what is really at stake? Where the Model Begins to Strain
Start with the most basic point. Artists are unable to say no. The model removes any opt-out. Even a small independent journalist, a writer, or a photographer will have to give their work to be used as an example in AI training. The payment is later with a pooled system. But copyright has never been merely a matter of money. It has never lacked an element of control. The freedom to determine the manner, place, and person by whom one wishes his or her work to be utilized. That aspect appears to be smaller here.
Is that loss compensated by compensation alone? It is not a yes that comes easily. Next, there is the issue of distribution. The plan relies heavily on copyright societies as well as a central collection body.
The creative economy in India is not well organised. A lot of creators fail to register their works. A lot of them do not belong to any formal society. A village folk singer or a freelance illustrator on the Internet may never get into this system. So where does their share go?
The risk is simple. The royalties are mostly given to those who are already visible and organised. Others are not within the circle. The paper does not clearly show how this gap would be addressed.
It also has a technical dimension that seems unexplored. The model is talking about legally available information. But what does scale do? AI systems learn on scale by using huge datasets extracted from the internet. It is not easy to trace the source of every piece of content. And when training is finished, the data is not held in a traceable and tidy manner. It is integrated into a greater system. So, how does one match output back to input? How do you measure use? And these are not petty problems. They lead to the core of whether the system is operable.
Looking Beyond India for Clues
The same problem is faced by other countries. In America, fair use is the point of contention. It is yet to be determined by the courts whether AI training is considered in it. The example of the New York Times controversy indicates that the solution is not so clear.[9]
The European Union has assumed a more moderate course. It allows text and data mining, but with an opt-out option. When a creator explicitly reserves his or her rights in a machine-readable manner, it must not be violated by AI developers.[10]
This is not the limit of the proposal made by India. It eliminates that option. That may speed up innovation. But it is also a tipping point.
Japan and Singapore have gone further and been more lenient by permitting the general use of the data in training.[11] Nevertheless, their systems work under varied market conditions, varied enforcement realities. The model of India is different. It is more centralised. More mandatory. And, perhaps, more taxing to faith.
The Administrative Crisis No One Can Avoid
Legal theory is not the most difficult aspect of this whole model. It is an administration.
To ensure that the system is effective, it should be able to recognize work, trace ownership, and make payments properly. However, in a large and diverse country such as India, it gets complicated quite fast. Numerous works are not documented. There can be ambiguity of ownership. Disputes are common. The burden of AI training is also added to this, which makes the task heavier.
The creators will not even be aware of whether their work was used without good transparency rules. In the absence of a well-defined claims process, they might not be aware of how to claim. And with no special quota assigned to unorganised makers, the system will fail to include the people who are most in need. It is possible to envision a scenario in which a small number of full-sized rights earners are making consistent revenues, and thousands of minor creators get nothing. That would be self-defeating. Then it is not fairness in design only, which is the real test. It is equity in performance.
A More Prudent Way Forward
The DPIIT paper should be credited with initiating this discussion.[12] It attempts to make order out of a disorderly place. And it does know that creators are not to be omitted. However, the existing model can be excessively biased.
A more refined solution would retain the original concept but smooth it out. At least some types of work can be regained by an opt-out system. Sharp disclosure regulations may render AI firms more responsible. A specific fund on unorganised creators might close distribution gaps. These are minor changes that can bring the balance of the system nearer.
Finally, it is not a question of whether India should facilitate the development of AI. It should. But it should also safeguard those whose labor lies behind that growth.
India does not have to decide between innovation and authorship. It requires a system that honours both. One that can keep the technology on the path, but not by stealing the rights of creators silently. Once the law starts to consider creative work as something that has to be shared at all times, then the meaning of ownership in itself can even begin to be eroded. And that is a change to consider, before it is too late.
[1]ANI Media Pvt. Ltd. V. OpenAI OPCO LLC, (Delhi High Court) (order reserved after hearings concluding March 27, 2026), reported in Business Standard (Apr. 2, 2026).
[2]Dep’t of Promotion of Indus & Internal Trade [DPIIT], Working Paper on Generative AI and Copyright (Part I) (Dec. 8, 2025)
[3]The Copyright Act, 1957, No. 14, Acts of Parliament, 1957 (India).
[4]Id. § 14.
[5]Id. § 51.
[6]Id. § 52.
[7]E. Book Co. v. D.B. Modak, (2008) 1 S.C.C. 1 (India)
[8]DPIIT, supra note 2
[9]N.Y. Times Co. v. Microsoft Corp., No. 1:23-cv-11195 (S.D.N.Y. filed Dec. 27, 2023)
[10]Directive 2019/790, of the European Parliament and of the Council of 17 April 2019 on Copyright and Related Rights in the Digital Single Market, art. 4(3), 2019 O.J. (L 130) 92
[11]Copyright Act, 1970, art. 30-4 (Japan) (exemption for exploitation not for the purpose of enjoying the thoughts or sentiments expressed in the work, including data analysis/AI training); Copyright Act 2021, §§ 243–244 (Sing.)
[12]DPIIT, supra note 2.
Author: Ms. Sneha Awasthi, student at Chhatrapati Shahu Ji Maharaj University, Kanpur
Link to similar articles: https://jpassociates.co.in/traditional-knowledge-digital-library-tkdl/
Link to DPIIT’s Working Paper on Generative AI and Copyright: https://www.dpiit.gov.in/static/uploads/2025/12/ff266bbeed10c48e3479c941484f3525.pdf