India recommends AI firms to pay creator royalties at rates set by govt panel

In the working paper titled ‘One Nation, One License, One Payment: Balancing AI Innovation and Copyright’, the committee led by the Department for Promotion of Industry and Internal Trade rejects voluntary licensing — deals between AI developers and individual companies, such as OpenAI’s content licensing agreement with the Associated Press — saying it would lead to high transaction costs, long negotiations and unequal bargaining power that disadvantage small creators and startups, and would fail to offer broad, dependable access to training data.

A committee member, speaking on condition of anonymity, said that they rejected the idea of voluntary licensing as it “only favours big players on both the AI side, and the copyright holders’ side”. “The report recommends AI companies have non-discriminatory access to copyrighted work, and takes care of rate fixation with an assurance that all copyright holders are compensated in an equitable way. We have made sure that not a few big companies corner royalties,” the committee member said.

It comes amid growing scepticism of news publishers in several jurisdictions, including in the United States and India, over concerns of copyrighted material, such as news reports, being used by companies like OpenAI for training their foundational models, without permission or payment. Here, creators include musicians, news publishers, and authors.

This has led to court cases, including in India, where publishers — members of the DNPA, including The Indian Express, among others — have mounted a legal challenge against OpenAI over the “unlawful utilisation of copyrighted material”.

However, some in the industry are raising red flags over the royalty rates fixed by a government-appointed panel. “This simply does not happen anywhere else in the world, so why should India be an exception? Companies should be allowed to discuss mutually beneficial licensing rates among themselves,” a senior executive who represents a key creative group said, requesting anonymity.

What model of compensation has the DPIIT-led committee recommended?

While rejecting direct, voluntary content licensing models, the committee has recommended a “hybrid model,” under which a mandatory blanket license with a statutory remuneration right for the creators and copyright holders would be established for the use of all lawfully accessed copyright-protected works in the training of AI systems.

The report states that lawful access by AI companies of copyrighted works will be a prerequisite, meaning that AI developers would not be allowed to rely on the mandatory license to bypass existing or future technological protection measures or to gain unauthorised access to works behind paywalls without making the necessary payment.

A new umbrella industry body, called the Copyright Royalties Collective for AI Training (CRCAT), will have to be set up by rights holders and be designated by the Central government under the Copyright Act, 1957. The CRCAT would serve as a centralised facilitator that streamlines collection of royalties for use of copyrighted content for training of AI systems, under a mandatory blanket license. Only organisations can be its members, and only one member per class of work would be allowed.

The report notes that since some unorganised sectors have no registered copyright societies or Collective Management Organisations (CMOs), CRCAT may not have membership from such sectors, but its base can be expanded over time to cover unorganised sectors.

How will royalty rates be determined?

The Centre will constitute a committee to determine the royalty rate, according to the report. It will consist of senior government officers, senior legal experts, financial or economic experts, and technical experts with expertise in emerging technologies. A CRCAT member from and AI developers’ representative will also be part of the committee, according to the recommendations of the report.

Based on inputs from stakeholders, market data, and economic analyses, the panel will set royalty rates, while ensuring that they are fair, predictable, and transparent. It will also review and adjust rates every 3 years, to reflect technological and market developments.

Notably, the rates set by the rate setting committee may be challenged before the court and would be subject to judicial review.

At this point “flat rates” seem to be the most appropriate model, according to the report. The committee can set a certain percentage of the gross global revenue (excluding taxes) earned by an AI developer from the commercialisation of the AI system trained on copyrighted content.

The report noted that government departments have established mechanisms to manage pricing and tariff matters in key sectors. For instance, it said, the Ministry of Railways reviews and proposes revisions to passenger fares and freight charges. The Department of Food & Public Distribution determines the Central Issue Prices (CIP) for essential food grains like rice, wheat etc. under the Public Distribution System (PDS). The Department of Fertilisers sets the Nutrient Based Subsidy (NBS) rates for phosphatic and potassic fertilisers.

Will the rates apply retrospectively?

AI companies have already ingested copyright protected works from several creators in their training systems. Due to this, the committee has recommended that the obligation to pay royalties on the basis of revenue percentage-based rates set by the government shall apply retroactively. In other words, AI developers who have already trained their AI systems on copyright protected content and are earning revenues by commercialising such AI Systems would be obliged to pay the prescribed royalties.

“Many AI Developers have built AI Systems which are commercially successful and generating huge revenues. In order to ensure fairness and accountability, such AI Developers must be required to pay royalties to copyright owners for past usage of their works. This is not a punitive measure, but a corrective mechanism to help create a balance in the creative ecosystem,” it said.

How will royalties be divided among copyright holders?

AI companies will be required to submit a declaration to the CRCAT containing a ‘Sufficiently Detailed Summary’ of the datasets they have used to train their models, as part of their transparency obligations. This declaration would include information including the category of data (text, images, music, etc.) used, source of data (social media platforms, hard copy/electronic publications, online libraries, public datasets, proprietary datasets etc.), and the nature of data.

The total royalty amount would then be apportioned proportionally by the CRCAT among different classes of works, based on the proportion of their usage in AI training. This would ensure that categories of works more heavily utilised (for example, audiovisual content or music or literary publications) receive a correspondingly appropriate share of the royalty pool.

In the event of a lawsuit against an AI developer for failing to offer the fixed royalties for the utilisation of copyrighted content, the law would work on the presumption that the claim of the content owner is valid (subject to the usual requirement to demonstrate a prima facie case) until proved otherwise by the AI developer.