Department for Promotion of Industry and Internal Trade (DPIIT) has proposed not allowing AI companies to train their models on copyrighted content for free, warning that this approach would erode incentives for human creativity and distort the value chain

The body has signalled support for a hybrid statutory licensing model that would let AI developers use any lawfully accessed copyrighted content for training, but only after paying royalties once their models become commercial

This recommendation sits at the core of DPIIT’s newly released working paper on AI and copyright, the first part of a policy exercise launched in April 2025 by an eight-member committee

In its working paper on GenAI and copyright, the Department for Promotion of Industry and Internal Trade (DPIIT) has proposed not allowing AI companies to train their models on copyrighted content for free, warning that this approach would erode incentives for human creativity and distort the value chain.

Instead, the body has signalled support for a hybrid statutory licensing model that would let AI developers use any lawfully accessed copyrighted content for training, but only after paying royalties once their models become commercial.

This recommendation sits at the core of DPIIT’s newly released working paper on AI and copyright, the first part of a policy exercise launched in April 2025 by an eight-member committee.

The core proposal is a hybrid statutory licensing model. Under this, AI developers would receive an automatic blanket licence to use all lawfully accessed copyrighted works, books, music, films, news, artworks and more, without needing prior permission or negotiating individual deals.

They would only pay royalties after their models begin earning revenue. A central government-appointed body would collect and distribute these royalties, while a separate committee would set rates, subject to judicial review.

DPIIT argues this model balances two goals: ensuring creators are paid fairly, and preventing bottlenecks that could block startups from accessing training data. It also aims to reduce bias by ensuring broad access to high-quality content, something voluntary or opt-out systems may restrict.

The committee said it chose a “no absolutes” approach after rejecting global models such as blanket text-and-data-mining exceptions and voluntary licensing systems, concluding these would not work for Indian needs.

The committee contends that giving AI firms free access to copyrighted works would amount to a zero-price licence that could polarise income across the AI value chain. It warned that if AI-generated outputs begin replacing human work without compensating creators for training use, it would weaken incentives to produce new content.

This “substitution effect”, the committee said, would harm India’s broader creative ecosystem in the long run.

But the approach comes with trade-offs. Creators would not have an opt-out or the right to refuse use of their work for AI training, effectively making this a compulsory licence covering the entire creator economy. Critics say such centralisation could create transparency risks, especially since the paper rejects mandatory disclosure of training datasets — a safeguard widely adopted in other jurisdictions.

The working paper acknowledges these concerns but says the hybrid model best meets India’s objectives: fair compensation, predictable rate-setting, minimal disputes and easier access for both big and small AI players.

The draft is now open for a 30-day public consultation, inviting feedback from creators, publishers, AI companies and the public before the government moves towards shaping India’s first dedicated legal framework for AI training and copyrighted content.

The Indian AI–Publisher Dispute Backdrop

This latest move from the government comes after a year of escalating disputes between India’s largest news publishers and global AI firms. In the past, multiple media houses including ANI, NDTV, Network18, The Indian Express and Hindustan Times, have accused OpenAI of using their copyrighted articles to train its large language models without permission, payment or licensing.

In parallel, the DPIIT has been running stakeholder consultationswhere publishers argued that AI training on their content without consent amounts to infringement, while tech companies maintained that large datasets, including copyrighted works, are essential to build competitive models.

Industry bodies such as the DNPA and FIP have also pushed for a regime that ensures fair compensation and enforces creators’ rights online.

The backdrop also includes a widening legal battle in the Delhi High Court, where news organisations, publishers and music labels have all sought protection against unlicensed scraping and reproduction of their works.