DPIIT Committee on Generative AI and Copyright: Working Paper Part 1, "One Nation One License One Payment" (December 2025)

India's Copyright Act, 1957 has no express text-and-data-mining exception and permits only limited fair dealing. The Government's India AI Governance Guidelines record that, under Section 52, the fair-dealing exceptions cover private or personal use including research, are restricted to non-commercial use and do not extend to organisational or institutional research, so "they may not cover many types of modern AI training".¹¹ Lawsuits such as ANI's against OpenAI in the Delhi High Court had made the question pressing.

On 28 April 2025 DPIIT issued an office memorandum forming a multi-stakeholder committee to study the intersection of AI and copyright and to examine whether the 1957 Act can accommodate AI-generated content. MediaNama reported that the committee is chaired by Himani Pande, Additional Secretary at DPIIT, with Simrat Kaur, Director of Copyright, Design and CIPAM, as member and convenor. The other members are Anurag Kumar of MeitY, Chockalingam M and Sudipto Bannerjee of NASSCOM, IP attorneys Ameet Datta and Adarsh Ramanujan, and Prof. Raman Mittal of the University of Delhi. Its terms of reference were to identify the legal and policy issues, examine the sufficiency of the Act, recommend changes where needed and publish a working paper.⁴

The India AI Governance Guidelines of November 2025 note that the DPIIT committee was formed in April 2025 and that it was examining the legality of using copyrighted work in AI training, the copyrightability of works produced by generative AI and international practice. The Guidelines' drafting committee suggested that the DPIIT committee "may consider a balanced approach" that enables text and data mining while protecting rights holders, and said it awaited the committee's detailed recommendations.¹¹

On 8 December 2025 the committee released Part 1 of its Working Paper on Generative AI and Copyright, titled "One Nation One License One Payment: Balancing AI Innovation and Copyright".¹²³⁷ Part 1 examines the use of copyrighted works as training data and leaves outputs to a later paper.⁷

As described by MediaNama and Bar & Bench, the paper proposes a hybrid model with these elements:²

• Mandatory blanket licence. Developers who lawfully access copyrighted material, by buying, licensing, subscribing to or otherwise legally obtaining it, may use it to train AI without negotiating individual permissions. Lawful access is the compliance boundary, and access rights are treated as distinct from copyright permission.
• No opt-out. Creators cannot block the use of their works in training and are compensated through a statutory right to remuneration.
• A central collecting body. The proposed Copyright Royalties Collective for AI Training (CRCAT) would collect payments from developers and distribute royalties to creators. Creators must register their works to be paid. Unregistered works could still be used for training, but their creators would not be paid.
• Royalty design. Developers would pay under pre-defined royalty formulas and revenue thresholds, and MediaNama reported that royalties would be a percentage of revenue from models trained on such content. MeitY's stated view, as recorded in the paper, is that royalties should be triggered only after a model or developer crosses a revenue threshold. The exact rates are not set out in the paper.
• Forward-looking scope. The framework does not address past unauthorised use or ongoing scraping disputes, which remain under current law and pending litigation.

The committee explained that the model would make works available "as a matter of right", provide fair compensation and reduce compliance costs for developers.⁶

The committee considered and rejected several alternatives, according to MediaNama's account of the paper:²⁶

On the existing Act, the committee said that amending Section 52(1)(a) to facilitate AI training would "neither help strike a balance, nor" effectively address developers' legal exposure. It reasoned that fair dealing is an exception and a defence, in contrast to the exclusive rights granted to owners by Section 14, so the burden of proof falls on the developer, who would also have to bring the use within one of the listed purposes such as research, criticism or review, or reporting current events. Even an amended provision would therefore not remove legal uncertainty.³

The paper notes that most technology and AI stakeholders consulted wanted a blanket TDM exception, but it chose licensing because an exception would deny creators the chance to seek compensation.⁶ It also says it "does not attempt to resolve" whether training infringes or whether fair dealing applies, and explains that case law "may take time", which is why a legislative solution is proposed while ANI v. OpenAI is pending.⁷

MeitY is reported to support the hybrid model, arguing that developers need broad datasets and that creators should be paid because outputs can replicate artistic styles.²

NASSCOM dissented in a 24-page submission. It asked for a TDM exception covering commercial and non-commercial AI training on lawfully accessed content, with a good-faith safeguard, machine-readable opt-outs for publicly accessible content and contractual reservations for non-public content. It described TDM as "at most a technical infringement" and argued that the licensing scheme adds administrative burden that could slow innovation and hurt smaller companies most.²⁶

The Business Software Alliance argued that training extracts non-copyrightable patterns and should need no permission once access is lawful, urged a statutory TDM exception on the model of Japan and Singapore, and said mandatory royalties and centralised licensing would add costs for smaller developers. It supported full remedies for infringing outputs and narrow protection against unauthorised digital replicas.⁵

In legal commentary, SpicyIP contributors published several critiques. One argued that the paper reverses the proper order of reasoning, recommending a licensing regime while leaving copyrightability, infringement and fair dealing unresolved, and that adopting a licence could freeze judicial development of the doctrine.⁷ Four SpicyIP contributors filed comments on 5 February 2026 objecting to the paper's premise, its treatment of the law and the practicalities of the scheme. Using the paper's own estimate of over a billion copyright owners, they illustrated that a royalty pool could yield very little per owner.⁸ MediaNama's analysis noted that the proposal shifts copyright from a right to exclude toward a right to remuneration, and raised open questions on valuation, distribution among creator types and whether payment alone sustains creative incentives.⁵

The sources reviewed do not state the comment deadline on Part 1; submissions were still being made in early February 2026.

Reported replies in Parliament show that the Government treated the subject as under review rather than settled. Bar & Bench reported on 17 December 2025 that the Centre told the Lok Sabha it was reviewing the adequacy of the Copyright Act, 1957 to address the legal challenges of generative AI, and that questions of authorship, ownership and copyrightability of AI-generated works were under examination and would be addressed in a white paper.⁹ MediaNama attributed the statement to the Minister of State for Commerce and Industry, Jitin Prasada, on 17 December.⁶

The India AI Governance Guidelines called for "targeted legislative amendments" to encourage innovation, naming copyright as one area, and awaited the DPIIT committee's detailed recommendations.¹¹

As of 15 October 2026, the sources reviewed show that Part 1 of the working paper remains a consultation document. No Copyright (Amendment) Bill implementing a licensing scheme or a TDM exception has been identified, and CRCAT has not been established.

Part 2, which is to address "copyrightability and authorship of GenAI-generated outputs, including the applicability of moral rights and attribution of liability", had not been published when SpicyIP wrote in February 2026.⁸ On 1 September 2026 MediaNama reported that it could not find a published Part 2, noting that Business Standard had reported in December 2025 that it was expected within about two months. The same article reported a Copyright Office order holding that an AI system cannot be the author of a work while the work itself can be protected, and observed that it was unclear whether that order settled the authorship question the committee had reserved for Part 2.¹⁰

In the meantime the courts continue to shape the law: MediaNama reported on 1 October 2026 that the Delhi High Court had declined an interim injunction to ANI in July 2026 and that the Division Bench also declined interim relief in September 2026 and asked OpenAI for a response.¹²

The full text of Part 1 could not be read for this entry, because the official DPIIT file returned an access error and the mirror was too large to retrieve. Descriptions of its contents rely on the news and commentary cited.