Even as the information age is debating one of its most contentious issues of the artificial intelligence (AI) era—that of due credit to the content used to train the large language models (LLMs) that train today’s AI tools, India has popped up with its own solution.
Even while it has been hailed as a long-overdue attempt to catch up with global AI policy, why is the proposal already ruffling feathers, at least from industry quarters?
The issue arises from media firms to research organisations and content creators feeling left out as the few ‘Big Tech’ firms that run the AI tools use their work to train their LLMs, spawning better and better AI services—and profit from it. While they get left out in the cold.
There is a sense of deja vu in it all, stretching from the dictatorship of the algorithm over the past decade or so, where search engines profited from ad analytics for peddling aggregated content created by others.
In fact, sensing a similar, perhaps even more potentially lethal scenario repeating itself, there are already umpteen cases in courts, the most famous of them being that of The New York Times filed against OpenAI (the parent of ChatGPT) and Microsoft, alleging that its copyrighted content was used without payment or permission to train AI models.
India’s proposal, actually made a few months ago, put down in a formal government white paper this week, suggests a hybrid statutory licensing system which will allow AI developers to use, what the document terms “all lawfully accessible” copyrighted content for training under a mandatory blanket license.
“The new proposal…encourages better record-keeping, cleaner licensed datasets, and standard ways of tracking where data comes from. This can reduce AI mistakes like hallucinations and make bias checks more accurate,” said Neelima Vobugari, co-founder and COO of the AI firm AiEnsured.
The working paper was prepared by an 8-member committee from the Department for Promotion of Industry and Internal Trade (DPIIT) of the Indian government and aims to ascertain how copyright laws can address the issues raised by the new generative AI wave sweeping the world. It proposes a royalty to be paid by AI firms to a copyright society for using legally accessed copyrighted material for training purposes, but doesn’t offer any solutions for free-use formats, instead hinting at a blanket permission structure.
The government proposal is to set up a Copyright Remuneration Collective for AI Training (CRCAT), which would receive payments from AI firms and distribute them to existing copyright societies or similar collective management organisations (CMOs) that creators could join.
The rule stipulates that payment needs to be made only when the AI tool is commercialised and not when the material is used to train LLMs. Payment can also be a retroactive obligation—meaning AI firms will have to pay even if the material was used for training at a previous period, with the compensation to be paid the moment the AI tool is commercialised.
Many are unhappy with the proposed norms. Global tech companies have come out saying the model is too complex and could slow India’s AI growth. One major firm even warned that the new norms, if they are put into effect, could kill India’s AI startup ecosystem.
“This risks opening a massive can of worms. The regulatory load will be enormous, and expecting a central agency to manage filings, verify claims, set royalties, and act as the middleman between creators and AI companies feels unmanageable,” said Pawan Prabhat, co-founder of the AI startup Shortfalls AI.
“Almost every decision will still be open to judicial review, so who will realistically take that risk?” he added, “AI is a global industry, and having India operate on a completely different rulebook could create more friction than clarity. The idea of the government becoming the exchange through which royalties flow raises even more questions: How will it regulate, track, and distribute payments at scale? What happens when disputes arise?”
Public consultations are open for suggestions till January 7, after which the committee is expected to release a second such paper on the topic.
