Copyright protection: Better framework is needed for AI training

In a recent working-paper on generative artificial intelligence (GenAI) and copyright, the Department for Promotion of Industry and Internal Trade has proposed 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. Under this proposal, AI developers will receive an automatic licence to use all lawfully accessed copyrighted works without prior permission or individual deals. A central government-appointed body will collect and distribute these royalties while a separate committee will set rates, subject to judicial review.

 

This body will set and collect royalties from AI companies, distributing proceeds among copyright holders. Coverage under this system will extend to unorganised sectors. Ratesetting will be through consultation with government officers, legal experts, financial experts, emerging technology specialists, and AI developers. Creators will not have an opt-out or the right to refuse use of their work for AI training. The paper also rejects mandatory disclosure of training datasets — a safeguard widely adopted in other jurisdictions. If adopted, this system would make India a global outlier in terms of AI copyright regulation. The paper has drawn criticism on different grounds from stakeholders, including tech companies and lobbies such as Nasscom, and from content creators and publishers. Most stakeholders support opt-out mechanisms for content creators. Tech companies and lobbies call enforced royalties a “tax on innovation”.  Meanwhile, content creators also fear undervaluation of premium content. Moreover, this process could lead to a situation where the owners of copyright for global content will be reluctant to license their copyright into India for fear of their content being forcibly inducted for training. This could have adverse long-term consequences with Indians not having access to premium content with negative implications for higher education among other things.

 

This paper comes at a time when there are disputes in multiple jurisdictions between AI companies and owners of copyright.  Disputes between India’s news publishers and global AI firms have also escalated with multiple media organisations accusing tech companies of using their copyrighted articles to train their large language models without permission, payment, or licensing. The paper appears to be well intentioned in certain respects in that it will not allow AI companies to train their models on copyrighted content for free, warning that this approach will erode incentives for human creativity and distort the value chain. But a model of forced licensing with royalties set by the government is not an ideal approach. Legislation must protect copyright for content creators and owners, but it must also be aligned with international trends. A legally acceptable system for remuneration for copyright holders is needed. Instead of forced licensing and setting copyright rates of payment, another way forward could be to lay out a legal framework where content copyright is protected with suggested model contracts that allow AI companies and content creators to negotiate their own rates. This would be a more pragmatic approach.