Shivanand Pandit
India has introduced a revolutionary proposal requiring AI developers to make statutory payments for using copyrighted works in model training. This initiative positions India as the first nation to pursue a compulsory licensing framework for generative AI, aiming to balance innovation with fair compensation for creators.
The Department for Promotion of Industry and Internal Trade (DPIIT) expert committee has recommended a ‘hybrid model’ that blends blanket licensing with statutory royalties. Under this framework, AI developers would automatically gain access to copyrighted material for training purposes, while creators would receive assured payments through a regulated royalty mechanism.
The proposal, detailed in a working paper released on December 8, 2025, represents a major policy step in governing generative AI and safeguarding copyright. The eight-member committee—led by Himani Pande, Additional Secretary at DPIIT—was tasked with examining whether India’s existing copyright laws adequately address AI-related uses of creative content. The committee clearly favoured creator remuneration and rejected industry demands for unrestricted text and data mining (TDM).
The working paper titled “One Nation, One License, One Payment: Balancing AI Innovation and Copyright” dismisses the idea of voluntary licensing—individual agreements between AI developers and companies like the Associated Press and OpenAI—arguing that such deals would involve high transaction costs, prolonged negotiations, and power imbalances that hurt small creators and start-ups. The committee concludes that voluntary arrangements cannot ensure broad, reliable access to training data. The working paper is now open for a 30-day public consultation.
Urgent need for clear rules
India’s draft framework proposing a statutory licensing regime for AI training represents a significant intervention in a central economic question of the digital age: who controls and benefits from creative value in the era of generative AI. If data is likened to oil, then the reservoirs are owned by millions of creators—writers, filmmakers, journalists, musicians, artists, and independent producers. Yet this creative output is currently harvested on a massive scale by global AI developers, often without consent, attribution, or compensation.
The draft recognises that creative work is not a free resource. It recommends a mandatory blanket licence that would allow AI developers to use lawfully accessible copyrighted material for training, provided they pay royalties set by a new body—the Copyright Royalties Collective for AI Training (CRCAT). This centralised system would collect and distribute payments not only to major rights holders, but also to smaller and unregistered creators who are typically excluded from conventional licensing structures. Crucially, royalty obligations would arise only when an AI model is commercialised, reducing the need for firms to negotiate thousands of permissions while ensuring creators share in the commercial gains from AI.
This approach diverges sharply from the TDM exemption advocated by industry groups such as Nasscom. TDM was crafted for limited research purposes, not for training commercial AI systems worth billions. Globally, disputes are already unfolding—The New York Times has sued OpenAI in the US, alleging unauthorised use of its archives, while ANI has filed a similar case in India. These legal battles underline that “publicly accessible” content should not automatically be treated as free raw material for large-scale AI training. Comparable debates have played out in the UK and the US, where existing doctrines were never intended to legitimise industrial-scale ingestion of copyrighted work.
India’s proposal is ambitious but not without unresolved issues. Retroactive application of royalties would be difficult: past training datasets may not be traceable, and determining liabilities could be contentious. Furthermore, granting AI firms a statutory licence should not imply unrestricted access to all content. Creators must retain a meaningful right to exclude their work if they so choose. To maintain simplicity in the scheme’s early stages, a flat royalty could be used initially, with a later transition to differentiated royalty pools—text, images, music, video—to more accurately reflect the value of different content types.
The DPIIT’s working paper arrives at a time when legal ambiguity around AI training is untenable. Generative models require vast amounts of copyrighted data, and clarity is needed to prevent future litigation from becoming the default mechanism for dispute resolution. The proposal consciously avoids two extremes: unrestricted free use, which risks undermining creative livelihoods, and granular permissions, which are impractical given the architecture of large AI systems and would disproportionately burden smaller developers.
Instead, the paper advocates a hybrid model: a statutory blanket licence for all lawfully accessed works, combined with a remuneration right triggered by commercial deployment. This structure offers legal certainty, reduces entry barriers for start-ups by avoiding upfront fees, and ensures that creators are compensated when their work materially contributes to commercial products. However, operational challenges loom. Applying the framework retrospectively could be complex, and concerns persist about making inclusion in training datasets mandatory. Critics argue this may infringe upon the property-like rights embedded in copyright law. Additionally, establishing a central entity to collect and distribute royalties will require careful governance to avoid distortions, arbitrary rate-setting, or administrative inefficiencies.
Industry reactions have been unsurprisingly mixed. Nasscom contends that mandatory royalty payments could hinder innovation, while creators view the proposal as a long-awaited acknowledgement of their role in powering AI systems. By circulating the paper for public feedback, the DPIIT has created an opportunity to bridge these opposing viewpoints. Ultimately, the framework’s viability will hinge on how effectively royalty calculations are designed, how robust the governance and transparency mechanisms are, and whether data usage can be tracked in a practical and reliable manner.
What is clear is that a structured legal framework is urgently required. The blanket licence model is less a perfect solution and more a pragmatic response to the realities of AI training, where datasets are vast, intertwined, and impossible to license piece by piece. Its effectiveness will hinge on how faithfully legislation translates these ideas into workable institutions.
The rapid advancements in large language models stem from two forces: constant improvements in machine learning techniques and the ever-expanding access to text, visuals, and multimedia used for training. AI developers have often asserted that online information should be free for such purposes, even though traditional reproduction of the same content would require permission and payment. This has intensified a global confrontation between AI firms and the industries whose content powers these models.
In this context, the DPIIT working paper’s solution—a mandatory licensing system allowing unrestricted scraping but requiring royalty payments based on the AI developer’s revenue derived from Indian content—is a practical attempt to bridge interests. The committee recognises that enforcing opt-outs individually is unrealistic for most creators, and that AI models typically generate new outputs rather than verbatim reproductions. While concerns remain about how royalties will be allocated—especially for small publishers producing fewer works than large media houses—the need for a compensation system is immediate. Courts across the world are still grappling with these issues, and relying on judicial outcomes alone would leave India without a cohesive strategy.
The white paper and the dissent notes from industry representatives together offer a starting point for a collaborative framework. A flawed regulatory structure can be improved; the absence of one only strengthens AI firms’ advantage in reshaping media, information ecosystems, and the broader creative economy.
