India is crafting a policy outline on how architects of artificial intelligence (AI) can use copyrighted material, a development that promises to impact AI giants and startups digesting vast amounts of data to spit out everything from quirky images to medical analyses.
Companies building commercial AI models must pay a fee to the source of copyrighted content, a government committee recommended. The Department for Promotion of Industry and Internal Trade (DPIIT) panel floated a mandatory licensing model for AI developers, with compulsory access to all “lawfully accessed copyrighted content” for model training without seeking individual permission from the creators; in turn, the developers would be required to pay royalties through a single, government-designated collecting body managed by rights holders.
“The entity will function as a centralized body for licensing and royalty distribution under the compulsory blanket licensing model,” it said.
The proposal also suggested that companies may be charged fees retroactively, and the fee itself would be a percentage of the revenue that a company generates through the said AI model. The DPIIT paper said the new model would lower compliance costs for AI startups, reduce litigation risk, and create a level playing field between large and small players. It would also open a steady revenue stream for creators, including from India’s vast informal creative sector.
The proposal argues that India requires a distinct regulatory structure that actively encourages AI innovation while ensuring creators are fairly compensated when their works are used to train AI systems. The move, described as a “one nation, one license, one payment” hybrid model, aims to fuel India’s ambitious 10,300 crore IndiaAI Mission while ensuring creators are compensated.
However, mandatory licensing and centralized royalty do not align with technical realities, said Kazim Rizvi, founding director of The Dialogue, a tech policy think tank. “Contemporary AI systems process billions of data points in iterative, non-deterministic ways that do not preserve work-level traceability. Expecting developers to identify, account for, or retrospectively reconstruct the contribution of individual copyrighted works is technically infeasible. A system premised on granular attribution, expansive repertoire completeness, or post-hoc compliance risks imposing burdens that would disproportionately harm startups and MSMEs, stifle research, and undermine India’s competitiveness in global AI development,” Rizvi said, adding the model’s potential retrospective application further opens a “worm-box” of uncertainty.
“Training corpora evolves continuously, model weights cannot be reverse-engineered to reveal underlying works, and global regulatory practice does not impose retroactive obligations of this nature. Introducing such liabilities would chill investment, create interoperability risks, and generate systemic disputes without delivering meaningful benefits to creators” he said.
“The adjudication of such a proposal is what’s concerning—how do you raise and settle disputes for an AI model trained five years ago, and the datasets that may have been used in it? All of this will need to be simplified, taken into technical consideration, and then worked upon,” Rizvi added.
A senior official with direct knowledge of the matter added that the proposals will likely receive a lot of feedback and discussions. “These proposals are what is expected to be shaped into India’s amended copyright law taking AI content and base models into account. Given its significance, a lengthy set of discussions are likely to be expected.”
The draft proposal, which will remain open for 30 days for feedback, is the government’s first formal policy outline in an area that has sparked intense global debate over the future of intellectual property. It comes in the wake of soaring AI adoption, mushrooming AI startups and conflicts over the use of copyrighted content by AI developers.
A DPIIT committee formed in April, after examining frameworks in the US, EU, UK and others, concluded that India cannot afford to wait for international litigation to settle before making its move.
The core legal dilemma the paper addresses is that AI training involves multiple acts of reproduction—from downloading and storing works to generating temporary copies—that raise clear infringement questions under India’s Copyright Act. To resolve this, the DPIIT suggested designating a non-profit royalty-collection body, comprising copyright societies and collective management organizations, that would collect payments from AI developers and distribute royalties to both its members and registered non-members.
Madhav Krishna, CEO and founder of Vahan.ai said while the intent to compensate creators is fair, blanket licensing will introduce “real friction” for AI developers in India. “Unlike Western companies that trained on vast internet datasets without such constraints, Indian developers will now face regulatory and financial hurdles from the outset,” he said.
While the move seems directionally right, a key side effect could be a slowdown in AI development – something that ultimately harms India’s ambitions to build and scale indigenous AI capabilities. We need a balanced approach that protects creators without stifling innovation, especially for early-stage startups,” Krishna said.
As per Nikhil Kurhe, CEO and co-founder of Finarkein, a data and AI infrastructure platform, AI companies operate in a gray zone when it comes to negotiating licenses, navigating copyright, and managing uncertain liabilities. India is also becoming the largest accessible AI market for both global and domestic players, with a rapidly growing base of active users.
“A blanket statutory license can create a predictable and transparent path forward for everyone to navigate these challenges. However, its implementation must be fair and scale-dependent; otherwise, it risks becoming an entry barrier for startups and further strengthening incumbents,” Kurhe said.
Not all startups, however, are against the proposal. Ankush Sabharwal, cofounder of homegrown agentic AI platform BharatGPT, said that it “should not be a disruptor, and will legitimize the use of datasets while giving publications a revenue source.”
“While it will comparatively make AI model development more expensive than elsewhere in the world, startups can pay a fixed fee to procure high-quality, verified datasets from credible sources. This would, in the long run, reduce incidental expenses, while reducing the percentage of queries on which generative AI applications will hallucinate. In the long run, this approach would make it beneficial for AI startups to make and sell commercial AI models to clients across industries such as healthcare or other critical sectors,” Sabharwal added.
Crucially, the policy paper rejects the tech industry’s push for a broad “text-and-data-mining” (TDM) exception that would permit AI training without payment, a stance the committee argues would “undermine copyright” and leave smaller artists powerless to seek compensation. The paper also discards opt-out models, citing the unreasonable technical and legal burden they place on creators.
Two senior officials told mint that the proposal is a build-up to amending India’s copyright law, and factoring AI into account—making it important for the proposal to reach a consensus among industry stakeholders.
“The objectives that the proposed framework is seeking to achieve in terms of protecting the rights of IP holders (by having a compensatory model) and at the same time ensuring mandatory licensed access to quality datasets for AI developers appears to be a balanced approach and may be a step in the right direction. However, what’s challenging is how the policy may be implemented in the long run, which would have multiple complications to iron out before arriving at a consensus. Further, IP holders who may not want to provide access to their works may feel let down by the process of proposed mandatory license,” said Shailendra Bhandare, partner at Khaitan & Co.
The proposal arrives as India, through the IndiaAI Mission, is actively positioning itself as a global AI hub, backing foundational model startups and expanding compute capacity. The government maintains that this licensing framework is vital for granting domestic AI firms transparent, legal access to rich, copyright-protected datasets, reducing barriers to innovation, and aligning with the goal of building sovereign AI capabilities.
Entrepreneur Bhavish Aggarwal’s Krutrim AI is currently building what it calls India’s first homegrown foundational LLM, trained extensively on Indic languages. Sarvam AI is another leading deeptech startup focused on developing high-performing, culturally fluent LLMs, including its OpenHathi Hindi model. The government-backed BharatGen Technology Foundation, established by an IIT Bombay consortium, is dedicated to constructing a multilingual LLM framework, while established players like Tech Mahindra are working on their in-house Indic model, Project Indus.
Krishna Kamal Palakaluri, senior director at Mirror Security said, “If India pairs transparent royalty mechanisms with tiered pricing and strong data provenance standards, we can avoid a ‘pay-to-play’ ecosystem while still protecting creators. Done well, this framework could become a global reference model for secure, responsible, and innovation friendly AI development.”
Furthermore, the framework must remain strictly scoped to training data and explicitly exclude user prompts, logs, or security telemetry, said Palakaluri.
The policy paper comes in the backdrop of escalating legal battles over AI. News agency ANI Media has sued OpenAI in the Delhi High Court, accusing the ChatGPT developer of unauthorized use of copyrighted articles to train its LLMs. This runs parallel to major global litigation, most notably The New York Times v. OpenAI and Microsoft in the US, alleging the unauthorized use of millions of articles, and the Authors Guild class action, where prominent writers contend their books were copied wholesale, as well as specific sectoral challenges from Getty Images and the Concord Music Group, all testing the limits of fair use and transformative works in the AI era.
The DPIIT paper argues that regulating the use of copyrighted works is an existential issue for the creative economy, noting that unregulated AI training risks long-term harm by potentially leading to the “underproduction of human-created works” and an eventual, culturally sterile, machine-generated content loop, concluding that “True public good lies in supporting innovation and human creativity together.” The framework
AI adoption is soaring in India, with OpenAI recently stating that India is its second-largest market after the US. The government argues that a licensing framework allows domestic AI firms access to rich, copyright-protected datasets, reduces barriers to innovation and aligns with India’s broader goal of building sovereign AI capabilities.
