What's a Fair Split for AI Licensing? (and the option to opt-out)

Generative (or really, regurgitive) AI systems have largely been trained on creative works by human creators, and some part of what these systems have been trained on are works under copyright that have been used without permission or payment. Cue the use of the term 'stolen.'

There's a movement to license works to train AI systems, with the idea that it's better for authors and illustrators and publishers to make some money on the deal, and to have AI training come to the legal side of things.

Two recent developments in this space to be aware of:

In November 2024, it was announced that HarperCollins made a deal to license some of their nonfiction works for AI training, letting authors opt-in or opt-out, and offering a 50%-50% split with the authors, meaning the publisher would keep 50% of the license fee.

One fascinating aspect of the deal was that HC wouldn't say initially which AI platform was licensing the materials. (The Authors Guild did a piece on that offer here. And Emarketer reported later in November that the HC deal was with Microsoft.)

Screenshot of Publishers Weekly article "Agents, Authors Question HarperCollins AI Deal"

And in December 2024, the Authors Guild put out this statement: AI Licensing for Authors: Who Owns the Rights and What's a Fair Split? stating that AI training is not included under "Subsidiary Rights" as in most current contracts, and that:

"a fair split for most deals will give 75–85 percent to the author, depending on the publisher’s role. Note that some academic presses are presently offering authors a mere 25 percent, which unfairly allows the publisher to reap almost all of the benefit while leaving most of the risk to the author. In all cases, the splits must be negotiated between authors (and their agents) and publishers as a new type of use that was not contemplated under existing contracts.

Most importantly, as these licenses are outside of the original publishing agreement, the author’s share of any publisher AI licensing deal should flow directly to the author and not be applied against an advance."

screenshot of the Authors Guild statement: AI Licensing for Authors: Who Owns the Rights and What's a Fair Split?

Certainly, an author or illustrator being able to opt-out is critical.

Also, if you do opt-in, once your work is part of the training of an AI system, it doesn't seem like there's a workable mechanism to get it back again and remove that information from the knowledge-base the AI has developed.

The metaphor in my mind is that you can't use eggs to bake a quiche and then once it's baked remove the eggs. The eggs are in there. Which leads me to question the timeframe of deals like HC's with Microsoft - they're saying it's just for three years. 

These are early days for AI, and I hope keeping up on these developments is helpful for you as well.

Because none of us know when our publishers might let us know there's been an offer to license our works for AI training, and these are things we should consider.

Illustrate, Translate, and Write On,
Lee