New Delhi’s proposal focuses on democratic AI governance with free data
By T N Ashok
In a dramatic shift in global AI policy, an Indian government panel has recommended that developers of large-language models and other generative AI tools pay royalties to creators whose copyrighted works — from news articles and books to music and films — are used to train AI systems.
The proposal, part of a sweeping working paper released this week, could reshape how companies such as OpenAI, Google (and by extension its AI arm, Google Gemini) operate in one of their fastest-growing markets.
The recommendation, part of the “One Nation, One Licence, One Payment” framework devised by a committee under Department for Promotion of Industry and Internal Trade (DPIIT), would make access to copyrighted works for AI training contingent on a blanket, statutory license — and accompanying royalty payments when the AI outputs are commercialized.
The panel argued that unrestricted “scraping” and ingestion of copyrighted works by AI firms risks eroding incentives for human creators. In a world of generative AI fed largely on uncredited content, the economic value is derived heavily from these creators — whether journalists, authors, artists, or musicians. Without remuneration, the committee warns, the creative ecosystem may suffer, especially in India’s diverse and rich — but often under-compensated — creative sector.
Moreover, the panel contends that a blanket-licence approach — rather than requiring individual deals with thousands of rights-holders — will simplify compliance and avoid unfair bargaining power. Under an alternative voluntary licensing model, smaller creators could be disadvantaged.
The royalty scheme isn’t just forward-looking: it would apply retroactively. AI companies that have already trained models on copyrighted works and are earning revenue would also be required to pay a share.
Crucially, the panel’s proposal does not set a fixed royalty rate yet. Instead, it calls for the government to appoint a rate-setting committee composed of senior officials, legal and technical experts, members of the rights-holders’ body (the proposed Copyright Royalties Collective for AI Training, or CRCAT), and representatives of AI developers. That committee will examine stakeholder input, market data, and economic analyses, then set a transparent, predictable royalty — likely as a percentage of global revenue generated by a commercial AI product trained on copyrighted content.
Until the rate is fixed, AI firms pay little or nothing — a status quo that the report disparages as unfair. That means currently there is effectively no royalty — the proposal aims to change that.
By making the licence mandatory and global (i.e. applying to all copyright-protected works lawfully accessible), the government aims both to prevent lock-outs by powerful rights holders, and to broaden compensation to even smaller or informal creators not part of established collective-management bodies.
India occupies an outsized and growing position in the global AI market. According to the committee’s working paper, India ranks among the top revenue streams for companies like OpenAI — second only to the United States — and may well become the largest market in the near future.
With millions of users, demand for local language support, and a growing ecosystem of AI-enabled services, India presents both opportunity and risk for global AI firms. The royalty proposal signals New Delhi’s intention to assert more control over the terms of that engagement — ensuring that value flows back to Indian creators and the creative ecosystem, rather than leaving monetization solely in the hands of foreign tech giants.
At the same time, the move supports domestic ambitions. India is pursuing its own AI development and may favour homegrown models trained on Indian content — enriching local linguistic, cultural, and contextual relevance.
On the global front, aside from OpenAI and Google, major players in generative AI include Microsoft (which supports some models), open-source projects, and various startups. In India, a surge of interest in building indigenous AI tools is underway — some aimed at regional languages, local content moderation, or domain-specific applications, reducing reliance on foreign-trained models.
Interestingly, the report does not single out any company; its mandate — and the payment obligations — would apply broadly to all AI developers using copyrighted Indian material in training. Thus, regardless of whether a model belongs to a global firm or a local startup, the royalty requirement would apply once the model is commercialized.
As for Microsoft Gemini (or other Microsoft-backed initiatives), they would fall under the same broad regulatory umbrella if they use copyrighted material for training. The proposal itself does not single out one firm; rather, it challenges the broader industry practice of relying on mass ingestion of public data without compensation.
Unsurprisingly, the response from the tech industry has been sharply critical. Nasscom — an influential industry body counting both Google and Microsoft as members — has formally dissented, calling the mandatory royalty a “tax or levy on innovation.”
Critics argue that the proposal could hamstring AI innovation in India by increasing compliance costs, dissuading smaller firms and startups, and imposing retroactive financial burdens on companies that have already invested heavily in model training. The argument is that AI development depends on large-scale data ingestion — and the pay-per-use or licensing model may not scale.
On the other hand, content-owners — including news publishers, authors, artists, and media houses — have long complained that their work was being used to train massive AI models without their permission and without compensation. For many, the policy offers long-awaited recognition of their intellectual property rights and a path to share in the profits generated by derivative AI systems.
If implemented, India will become — to the best of public knowledge — the only major economy to enforce a statutory licensing and royalty-sharing regime for AI training on copyrighted content. That alone could transform global AI economics.
For creators: the promise of a steady royalty stream via a centralised body (CRCAT) could re-energise content production, particularly among smaller or informal creators who have often lacked bargaining power or visibility.
For AI firms: retroactive payments and ongoing royalty obligations could materially impact their business models. Large firms may pass costs on to users, raise subscription prices, or restructure how they ingest and license content — perhaps favouring licensed or in-house content over broad web scraping.
For Indian AI ambitions: the regulation could tip the balance towards domestically developed AI models, trained with licensed Indian content — better suited for local languages and contexts. It may also encourage collaborations between AI firms and Indian publishers/creators, possibly leading to an emergent licensing + innovation ecosystem.
The government has opened a 30-day consultation window for public and industry feedback before the proposal is finalized.
Given the stakes, expect significant pressure from global tech firms, industry bodies, and possibly foreign trade allies — arguing that heavy regulation will stifle innovation and raise barriers to entry. On the other side, expect continued lobbying from creators, publishers, artists, and media houses insisting on fair compensation and copyright recognition for works fuelling AI.
Moreover, questions remain: How high will royalty rates be set? Will retroactive charges discourage companies from operating in India? Will smaller startups and open-source efforts be disproportionately burdened? And will the central body (CRCAT) effectively ensure fair distribution to rights holders across India’s vast and often informal creative sector?
In short, India’s proposal — whether adopted as is or in a watered-down form — signals a new direction in global AI governance: from treating public data as a free-for-all to recognising data as a monetizable, creator-owned asset with real economic value. (IPA Service)
