Govt AI panel backs blanket copyright license and royalties for GenAI training

New Delhi: A government committee set up by the Department for Promotion of Industry and Internal Trade (DPIIT) has proposed a new “One Nation One License One Payment” framework to govern how generative AI systems train on copyright-protected works, pitching a hybrid model that combines a mandatory blanket license with statutory remuneration for creators.

In an executive summary released as Part I of its working paper on AI and copyright, the committee said the central challenge is to protect rights in human-created works “without stifling technological advancement” and calls for a balanced regulatory architecture that preserves the integrity of India’s creative ecosystem while supporting AI innovation.

DPIIT constituted the committee on April 28, 2025, to identify legal issues raised by AI systems, review the existing framework and recommend changes, with a specific mandate to examine both the use of copyright-protected works as training data and the copyright status of AI-generated outputs. The current paper addresses the training-data question; a second part will deal with authorship, moral rights and liability for infringing AI outputs.

TDM exception vs licensing: committee sides with creators

The panel held separate consultations with technology companies and with content-industry bodies. According to the paper, most tech and AI stakeholders argued for a blanket text and data mining (TDM) exception that would allow GenAI training on copyrighted works, while a smaller group backed a TDM exception with an opt-out right for rights holders. Content-industry representatives, by contrast, unanimously pushed for a voluntary licensing model.

After studying regimes in the US, Japan, the UK, Singapore and the European Union, and noting ongoing litigation in India and overseas, the committee concluded that waiting for court outcomes “may not be optimal” and that all the major models it examined – voluntary licensing, extended collective licensing, traditional statutory licensing and both forms of TDM exception – face “significant suitability challenges” for India if applied in their traditional form.

On the tech industry’s preferred TDM exception, the panel is unambiguous. Allowing such an exception for commercial AI training “would undermine copyright” and leave human creators “powerless to seek compensation” for use of their works, a result it calls especially unwise for a country with India’s cultural heritage and growing content industry.

Even a TDM model with an opt-out right was found to be inadequate. The paper warns that it would leave small creators largely unprotected because they may lack awareness, bargaining power or tools to monitor whether their content is scraped despite opting out. It also notes that once data is stripped of metadata and transformed, control is “irrecoverable”, and that widespread opt-outs could shrink training datasets, hurt AI quality and shift burdens from users to creators.

Hybrid model: Mandatory license plus revenue share

To reconcile access and compensation, the committee proposes a hybrid model anchored in a mandatory blanket license introduced via amendments to the Copyright Act, 1957. This would ensure that no copyright owner can withhold their works from use in AI training, provided AI developers have lawfully accessed the content.

Under this framework, all lawfully accessed copyright-protected works would be available for training as a matter of right, without individual negotiations. The model aims to cut transaction and compliance costs for AI developers, guarantee fair compensation for rights holders, provide judicial review over royalty rates and create a level playing field for start-ups as well as large players.

Royalties would be tied to the revenue generated by AI systems trained on copyrighted content, with a “certain percentage” of that revenue payable to rights holders. The royalty rates would be set by a government-appointed committee. The paper emphasises that payment obligations arise only once an AI system is commercialised, meaning no upfront fees during the training phase, a design choice intended to support early-stage innovation and start-ups.

CRCAT: A new central collecting body

To administer the regime, the paper proposes setting up a single umbrella collecting entity called the Copyright Royalties Collective for AI Training (CRCAT). It would be a nonprofit organisation formed by associations of rights holders and designated by the Central Government under the Copyright Act.

Only organisations would be CRCAT members, with one member per class of works, such as existing copyright societies or other collective management organisations (CMOs). These members would manage online “Works Database for AI training royalties” systems where any copyright owner, member or not, could register works with key metadata so that royalties can be correctly allocated.

Royalties collected from AI developers would flow into CRCAT and be distributed by these member organisations to both their members and non-members who have registered their works. Unclaimed royalties for sectors without CMOs would be held for three years; if no organisation emerges in that period, the money would move into a CRCAT welfare fund to support creators in those sectors, including through community studios, subsidised equipment and training programmes.

The paper acknowledges that CRCAT members will need to invest in capacity-building, digital infrastructure and governance reforms, but argues that implementation challenges should not be a reason to “eliminate the creators’ right to receive royalties”. Instead, it calls for robust distribution systems and a glide path that allows CMOs to build these capabilities.

Lawful access and legal certainty

Crucially, the proposed license is conditional on lawful access. AI developers would not be allowed to rely on the mandatory license to bypass technological protection measures or paywalls. Once they obtain access and pay any access fees, they can freely use those works for training without further permission. The lawful-access requirement is intended to operate prospectively; past activities will continue to be assessed under existing law and court interpretations.

The committee argued that this hybrid model offers “legal certainty” for AI developers, predictable royalty flows for creators and a sustainable infrastructure that mitigates AI bias and hallucinations by keeping broad, representative datasets available for training.

Concluding its assessment, the panel warns that unlicensed use of copyrighted content for AI training raises “legitimate copyright-related concerns” and that a blanket exception would lean too heavily towards AI developers, risking under-production of human-generated content and weakening cultural evolution. 

The proposed hybrid model, it said, aims to secure broad data access while preserving incentives for creators and reducing litigation risk as India navigates the next phase of AI growth.