The short
- A DPIIT committee has released a working paper titled “One Nation, One Licence, One Payment: Balancing AI Innovation and Copyright.” It suggests a mandatory blanket licence that lets AI developers train on all lawfully accessed protected works in exchange for royalties paid into a central pool.
- Rates would be set by a government-empanelled body, not by one-to-one negotiation. Creators and rights holders would gain a statutory right to remuneration but lose the power to refuse training use altogether.
- For Indian startups, the idea promises cleaner legal ground and fewer takedown threats: pay the levy, show compliance, train on broad Indian datasets without chasing individual signatures.
- For creators, the upside is fresh income streams and a structured way to get paid when AI systems rely on their work. The downside: worries that rates end up low, opaque or skewed towards bigger players.
- For global giants, the scheme looks like a territorial AI tax layered on top of existing compliance, and it may clash with “fair use” and opt-out regimes in other regions.
What is actually on the table?
The committee’s paper does not ban AI training on copyrighted Indian works. Instead, it treats such use as allowed—but only through a compulsory licence that comes with fees and rules.
In plain language, the draft regime would:
- Create a single national licence that covers books, journalism, films, audio, images and other protected content, as long as the AI developer accessed them lawfully.
- Replace case-by-case deals with centrally set tariffs, a bit like statutory music licences or compulsory radio royalties.
- Channel payments into one or more collection societies or a similar entity, which would then distribute money to authors, publishers, studios and other rights holders under agreed formulas.
- Remove “consent” as a control point for training: creators would not be able to say “no” to training use, only to argue about how the money gets split or how uses are tracked.
Crucially, this is still a working paper. Stakeholders have about a month to file responses. Parliament has not voted; no notification has been issued. But as a signal of direction, it is loud.
Why small labs and startups might cheer
Today, any Indian company that trains or fine-tunes models on scraped content lives in a legal grey zone. Global firms argue that training is a technical use covered by “fair use” or text-and-data-mining exceptions; many creators disagree; courts have not settled the question.
A blanket licence does three things for smaller builders:
- Predictable risk: pay the fee, follow reporting rules and you are broadly safe from sudden takedown notices for training runs that used Indian works.
- Cheaper access to scale: instead of paying premium rates to a few closed data vendors, labs could train on large Indian corpora—regional languages, local news, entertainment—under one scheme.
- Better fund-raising story: investors like clean compliance. “We are covered under the national AI licence” is easier to explain than “we scraped a lot and hope it’s fine.”
In effect, the state is saying: “We will sell you a legal umbrella. Use it, and stop pretending it is not raining lawsuits elsewhere.”
What creators stand to gain—and worry about
On the other side, the proposal tries to answer a basic grievance: content that cost years of effort is feeding US- and EU-built systems without clear payment or credit.
The upside for authors, journalists, studios and labels:
- New revenue streams beyond ads, subscriptions and syndication, especially for large back catalogues that still have training value.
- Less litigation stress: instead of chasing individual AI firms across borders, rights holders would focus on influencing national tariffs and splits.
- Stronger voice in data debates: once payouts exist, creators’ groups gain an obvious seat at any “IndiaAI” table that decides rates and scope.
The anxieties are equally clear:
- Low per-work payouts: if the pool is large but divided across millions of works, each book or article could see tiny returns.
- Opaque allocation: without transparent reporting and independent audits, smaller creators fear their share will vanish into admin and top-heavy splits.
- No veto power: some may simply not want their work used in training at all; the scheme gives them money, not that choice.
In other words: the regime promises fairness at aggregate level, but may feel blunt at the level of a single poem or independent film.
Unanswered questions for anyone building with AI in India
Even if you like the idea in principle, the details matter far more than the slogan on the cover page. A few unresolved questions:
- Scope: does the licence cover only training, or also synthetic output that closely copies specific works? Where does “inspiration” turn into “derivative”?
- Territory: does the obligation apply to any AI used in India, any AI trained inside India, or any AI that touches Indian works, even if the training run happened on foreign servers?
- Rate setting: will tariffs be flat, tied to compute usage, tied to revenue, or some hybrid? Are there discounts for research, open-source models or non-profit labs?
- Reporting: what logs must teams keep—datasets, checkpoints, prompts, only broad summaries? How intrusive would audits be for small firms?
- Compatibility with global regimes: how does a compulsory Indian licence interact with US fair-use arguments, EU opt-out lists and Japanese “train on anything” rules?
Until those edges sharpen, the scheme is an outline, not a full contract.
How to think about it if you run an AI product in India
If you are a founder, counsel, or lead engineer, you do not need to become a copyright scholar overnight. You do need a crisp internal stance. A practical way to frame it:
- Inventory first: write down what data you use today—open licences, proprietary feeds, scraped sources, customer uploads.
- Classify risk: separate clearly licensed or in-house material from grey-zone scraped sets that include Indian books, news, cinema or songs.
- Budget for a levy: assume some portion of future spend goes to a national licence, whether you like it or not. Treat it like cloud bills: part of the stack.
- Engage, don’t just complain: trade bodies, developer groups and creator collectives will all lobby on rates and rules. If you ignore that stage, you live with whatever others negotiate.
Most importantly, resist the instinct to assume “nothing will happen.” India rarely writes a working paper like this for fun; it signals where the Overton window now sits.
Rule — a simple mental checkpoint for this reform
A compact rule of thumb for Indian AI builders:
“If your system learns from Indian copyrighted works, plan as if a national licence and royalty tab already exist—and treat today’s lower-friction world as a temporary discount, not a right.”
That mindset won’t answer every legal question. It will, however, keep you from making long-term bets on a loophole that policy is explicitly trying to close.
Disclaimer
This bataSutra article is an editorial explanation of a draft policy document and is intended for general information only. It is not legal advice and does not create any lawyer-client relationship. Companies, investors, creators and other stakeholders should consult qualified legal counsel and review the full official text of the working paper and any future legislation or regulations before making decisions.