Is Compulsory Licensing for AI the Future?

On Monday in India, the Department for Promotion of Industry and Internal Trade published a working paper calling for a mandatory blanket license for AI developers. 

The paper, entitled One Nation One Licence One Payment: Balancing AI Innovation and Copyright, calls for a mandatory blanket license to allow AI training on lawfully accessed works. Rightsholders would not be able to opt out of this system.

In return, a series of collection societies (one for each type of work) would collect a government-mandated royalty rate. That rate would be based on a fixed percentage of the AI-generated revenue.  

Almost at the exact same time, the ​​EU Parliament’s Policy Department for Justice, Civil Liberties and Institutional Affairs released a report it commissioned entitled The Economics of Copyright & AI. 

It also called for a compulsory licensing scheme. Though it acknowledged that a broad exemption to copyright law could harm incentives to create, it also said that voluntary licensing was too onerous and would not be cost-effective.

To be clear, neither of these reports is binding, and the governments may set them aside and pursue other approaches. 

But they’re also impossible to ignore. The EU and India together represent nearly 2 billion people and account for over 20% of the world’s population. Any action they take with regard to AI and copyright will have a significant impact on not just their citizens but the world as a whole.

So why the new push toward a compulsory license? There’s one very simple reason behind it.

A Third Option in the AI Debate

Right now, when it comes to AI and how it is trained on human-created content, there are two diametrically opposed viewpoints.

The first argues that copyright law does, or at least should, bar AI from being trained on human-created work without a license. They feel that AI companies should get opt-in permission to use human-created and copyright-protected content.

The second side feels that AI companies either are or should be free to use whatever legally-accessible (and sometimes dubiously legal) content they can to train AI systems. They argue that such use is either already exempted under the current law, for example, a fair use, or that it should be if it isn’t.

We’ve seen arguments and decisions in both directions. In May 2025, the United States Copyright Office made it clear it didn’t feel AI training was protected under fair use. However, in January 2024, the Association of Research Libraries posted its stance stating the opposite.

This is why a compulsory license is appealing to so many in government. Taking either stance means risking the wrath of the other side. Side with AI companies and you’re angering an industry many believe is propping up the economy. Anger creators and artists, and that’s harming not only a large number of voters, but also activists who have proven they can rally around key causes.

A compulsory license is a middle ground. AI companies get their blanket license, but they have to pay for it. Human creators can’t say no, but at least they are compensated. Nobody gets what they want, but nobody completely loses.

However, many questions remain about how this would work.

How Would a Compulsory License Work?

Compulsory licenses aren’t anything new. In the United States, the best example is the Copyright Royalty Board (CRB). As per the Copyright Royalty and Distribution Reform Act of 2004, the CRB sets rates for various statutory licenses. 

For music compositions, the CRB controls the mechanical license, which determines how much interactive streamers (like Spotify and Apple Music) pay per stream and how much when a composition is used on a physical good or digital download. The latter comes up a great deal when artists release cover songs on their CD, meaning they used the composition but not the original recording.

For sound recordings, the CRB handles the rates for non-interactive streamers, such as Pandora (free tier) and SiriusXM. In all

In all these cases, artists can’t refuse the licenses and are paid what the CRB says is fair. 

The goal of these proposals is to create something akin to that for AI. However, there is a significant problem actually executing such a plan. The music industry is much more consolidated than the internet at large.

Current compulsory licenses target a single license and a single type of work. Things get much more difficult when you start trying to apply that formula to everything from books to movies and hashing out all the rights an AI company would likely need.

To make matters worse, the music industry is siloed. Three music publishers, Warner, Sony and Universal, control the lion’s share of the music publishing market in the US. Not coincidentally, those three names are also the largest record labels. Though there are a large number of independent labels, Merlin represents over 30,000 of them. 

The result of this is that it’s relatively easy to license the vast majority of popular music. With the internet, you would have to deal with every single website, YouTube channel, podcast and so forth. Then you would have to resolve ownership disputes and what qualifies for a royalty (including whether the content is AI-generated).

You have to do all that before paying a single cent in royalties. And that would then cover one country.

It’s a Sisyphean task and, ultimately, unlikely to enrich human creators significantly.

Bottom Line

While I understand governments wanting a third, better option, this is not the right path.

Simply put, such a system would be massive and unwieldy. Given how many issues we already have with the “streamlined” systems in the music industry, a new system that is many times larger and more complex would only yield more frustration.

But, even if it did work, it would likely do very little to reward creators. AI systems have been trained on billions and billions of works of all types. Apple, by contrast, earned over $400 billion in revenue in 2025. Very little of that comes from AI. 

But even if it were all AI, we’d still only be talking about a small pittance for each work. A 5% royalty, for example, would be $20 billion. Once you divide that among all creators, it amounts to very little per person or per work.

However, there is another problem. AI companies are not profitable, at least not yet. Though the proposals would base the royalty on revenue, not profit, AI companies are going to fight against paying any royalty, no matter how small, while they are still hemorrhaging money.

In short, this approach isn’t likely to work. But, if it does, it won’t help human creators, and AI companies won’t like it either. Sometimes an effort to find a middle ground helps no one, and that seems to be the case with these proposals.