Suno has formally responded to the lawsuit filed against it by independent artist Anthony Justice, who accuses the music AI company of copyright infringement. It wants the court to immediately dismiss one element of his legal action.
Like many of the copyright lawsuits filed against AI companies, Justice accuses Suno of infringing his copyrights twice. First on the input, when it copies his music into a training dataset. And then on the output, when the Suno AI generates tracks that are, in copyright terms, ‘derivative works’ of his originals.
Suno will use the fair use defence on the input, but that’s for another day. In a new court filing it focuses on the claims relating to the output. On that, it argues, Justice misrepresents how its AI works.
The only way a new recording outputted by the Suno AI “can infringe the rights in a pre-existing one”, it writes, “is if it borrows the actual sounds of the original”. And that, it insists, “is not how Suno’s tool works at all”, adding that Justice already knows this.
In other words, Suno’s claim is that it doesn’t generate new music by sampling and stitching together elements of songs or recordings in its training dataset. Instead, its AI analyses the music it has been trained on, and learns how to compose and produce new works.
“No Suno output contains anything like a ‘sample’ from a recording in the training set”, the new filing goes on, “so no Suno output can infringe the rights in anything in the training set, as a matter of law”.
Suno also notes that in many other legal battles between copyright owners and AI companies where allegations have been made about both the input and the output, the courts have generally dismissed the output claims, putting the focus entirely on the input and fair use debate.
“Every court to have confronted this issue”, says Suno’s filing, “has concluded that just because outputs of an AI model are in the plain-English sense ‘derived’ from a set of copyrighted works does not mean that the outputs of the model are all prima facie infringing ‘derivative works’ in the copyright law sense”.
So, while under US law permission may well be needed to create a ‘traditional’ human-made derivative work that samples and/or directly adapts an earlier work, that’s not what happens with generative AI, which is why courts have generally thrown out copyright infringement claims relating to the output.
Suno, of course, has also been sued by the major record companies. However, Justice’s lawsuit – which seeks class action status, with all US independent recording artists in the possible class – is important, because the majors only control a minority of the music Suno has likely copied as part of its training processes.
Plus the AI company could as yet settle with the majors and agree licensing deals, risking a two-tier system where major label artists’ recordings are licensed for AI training, with revenue potentially flowing back from Suno via the majors to music creators, while indie labels and artists are cut out.
It’s Justice’s claims on the input that really matter. In his lawsuit, the musician pre-empted Suno employing the fair use defence. Suno will argue that – under US law – AI training is fair use and therefore it does not need permission from creators or rightsholders when making use of existing content.
Justice’s lawsuit cited the report on copyright and AI published by the US Copyright Office earlier this year which, he said, stated that “the fair use doctrine does not excuse unauthorised training on expressive works”, such as music, “particularly when those works are used to generate substitutional outputs that may replace the originals in the relevant marketplace”.
Since Justice filed his lawsuit, there have been two rulings in AI copyright cases involving Meta and Anthropic which both concluded that AI training is in fact fair use. Although they both related to the written word rather than musical works, which might be treated differently.
Plus, in the Anthropic case, the judge was clear that the fair use defence only applied when an AI company copies training data from legitimate sources. It’s thought Suno’s training data came from illegitimate sources, with the AI company just scraping millions of tracks off the internet.
In its new filing, Suno notes that Justice’s copyright claims on the input replicate the major labels’ lawsuit. Both the AI company and the musician will “have ample opportunity to litigate that core issue”, it adds, but “this case should not be cluttered with an additional theory of liability that simply doesn’t track the law”.
Which means, it concludes, the court should dismiss Justice’s claims relating to the output.
