If one creates a copy of a copyrighted work for a licenced purpose can one reuse the copy for another purpose?

Question

Multiple lawsuits have been filed relating to training of generative AI models on copyright protected works. A recent suit seems to focus particularly on this aspect. It seems there could also be a case that the output is infringing (the neural nets were "overfitted" ... The labels see this as straight-up infringement), but please ignore that possibility for the purposes of this question.

In legal action coordinated by industry body the Recording Industry Association of America, Sony, Warner Brothers, Universal, and others have filed suit against Uncharted Labs [PDF], which develops Udio, and Suno [PDF] in New York and Massachusetts respectively.

Both cases focus on the same fundamental charge, that the upstarts used copyrighted music without permission to train their neural networks. It's a similar claim that has been filed against OpenAI, who has been accused of using news articles and other sources to train ChatGPT without explicit permission.

My understanding of this argument is that while Udio and Suno may have had a licence to listen to the music they did not have a licence to train an AI with the music. Both listening to the music and training an AI require the creation of an in memory copy of the music, and it is the creation of the copy for AI training that is the unlicensed activity that is infringing the copyright of the plaintiffs.

If we consider the hypothetical where this company wrote their own software in a low level language like C. This includes their music player and their training algorithm, both tools use the same memory location for the music, and all data is provided to the two algorithms by pointer. In this way the developers are certain that the training process never creates a separate copy of the work, it only ever reuses the copy that was created for the developer to listen to music.

Assuming the developer admits that they created the software for this exact purpose, and the plaintiffs accept the defendants were licenced to create a copy to listen to and there is no copy is created by the unlicensed training process, would any infringement have taken place in the training process?

Answer 1

In a comment, you ask

Copyright law restricts certain things, like making a copy. If you do not do one of those things how are you in violation of copyright law?

A license is permission to make a copy (or to do something else that is protected by copyright law); typically this permission is granted subject to certain conditions or restrictions. If you violate the conditions or exceed the restrictions then you have invalidated the permission you were granted.

Answer 2

"The developer admits that they created the software for this exact purpose" — "that purpose" being to use the copy they put into memory in order to train.

That being the case, they can no longer say that they are making the copy solely to listen to it. When they copy the song into memory they are also doing so with a purpose outside of the terms of the licence. And then they go ahead and actually do those things.

So, taking your facts as follows:

• Y holds the copyright to the song
• in order to get the song into memory, X had to make a copy
• Y licenced X to copy the song into memory only for the purpose of listening to it
• X copied the song into memory with a purpose of training on it and trained on it

then the copy was not licenced, and would be a prima facie infringement (subject to a fair use or fair dealing defence).

Answer 3

Headline question answer: no, Body question answer: yes

the instant you step outside your licence agreement, the copy is in violation of copyright law.