India is moving closer to changing how AI companies use copyrighted content, with different government departments backing a plan that would require developers to share revenues with creators.
The Department for Promotion of Industry and Internal Trade (DPIIT) has released the first part of a working paper proposing a statutory licensing system for AI training. The idea is straightforward: developers would be allowed to train AI models on copyrighted material, but once those models start generating commercial returns, the right holders would be entitled to a share of the revenue. The proposal has received backing from the Ministry of Electronics and Information Technology (MeitY), strengthening the case for a coordinated government approach.
How the licensing model would work
Unlike relying on broad “fair use” interpretations or opt-out mechanisms for text and data mining, the DPIIT paper outlines a mandatory blanket licence. Under this structure, developers could train generative AI systems on lawfully accessed copyrighted works without negotiating individual permissions upfront.
Compensation would kick in later. Once an AI product becomes commercially viable, royalties would be paid to the right’s holders through a central body, tentatively named the Copyright Royalties Collective for AI Training (CRCAT). This entity would be responsible for collecting and distributing payments across creators, publishers, and other copyright owners.
Officials argue the model is designed to avoid bottlenecks. Developers would not be forced into case-by-case negotiations, while creators would still retain a legally protected claim over how their work contributes to AI-driven revenue.
Read More | IndiaAI Mission Launches Free AI Course for Youth and Citizens Across India
MeitY’s push for balance
MeitY has endorsed the framework while calling for guardrails that reflect the realities of India’s startup ecosystem. In its comments, the ministry highlighted that access to large and diverse datasets remains essential for building competitive AI systems, particularly in Indian languages and local contexts.
At the same time, it underlined the need for “proportionate compensation” and suggested that CRCAT introduce a minimum revenue threshold. Such a threshold would delay mandatory revenue sharing until an AI product reaches a defined level of income, reducing compliance pressure on early-stage companies. The ministry also pointed to governance concerns, recommending a transparent and participatory structure for CRCAT to prevent disputes over how royalties are calculated and distributed.
Government alignment, with some dissent
In its formal submission, MeitY described the hybrid licensing approach as one that could “equitably meet” the twin goals of supporting technological development and protecting creative labour. The endorsement signals rare alignment across departments on a politically and commercially sensitive issue.
DPIIT has indicated that a second working paper will follow, suggesting the framework is still evolving and open to refinement based on industry and public feedback.
Not everyone is convinced. Nasscom, representing a large section of India’s technology industry, has recorded a dissenting view within the DPIIT paper. The industry body has argued that an opt-out system for rights holders, combined with a text and data mining exception, would offer greater flexibility and reduce legal uncertainty for developers.
Read More | Japan Toyoake Town Wants Residents to Limit Phones to 2 Hours Daily – But Will Anyone Listen?
Why this debate matters
In markets such as the United States, training AI models on publicly available data is often defended under fair-use doctrines. India appears to be charting a different course, one that explicitly links AI growth to creator compensation. That choice reflects the scale of the country’s creative economy and the speed at which AI tools are being adopted across sectors. It also raises difficult questions around enforcement, valuation, and oversight—issues that the next phase of consultations will need to address.
