AI & the music industry: “fair use” arguments take on a new tune in the US (via Passle) (2024)

In perhaps a natural next step following the slew of lawsuits filed in the photography, news and publishing industries against training so-called ‘Large Language Models’ using copyright material, certain music industry majors have joined forces against music-generation software developers Suno, Inc and Uncharted Labs, Inc in two US lawsuits.

Major music labels UMG Recordings, Inc, Sony Music Entertainment and Warner Records, Inc are among the plaintiffs in the cases against the developers, alleging that Suno Inc’s “Suno” and Uncharted Labs’ “Udio” AI-based, large language models were, without permission, trained on a large volume of copyright-protected sound recordings which are exclusively owned or controlled by the plaintiff labels.

Suno, Inc proffers itself on its website as “building a future where anyone can make great music… no instrument needed, just imagination”, while Udio is advertised as being positioned to “enable the next generation of music creators”. Users of both platforms can create songs which produce professional-sounding vocals and instruments by entering word prompts into the software.

Raging against the machines

The federal lawsuits against both Suno, Inc and Uncharted Labs, filed in Massachusetts and New York respectively, state that the plaintiff labels “have developed their enviable catalogs by discovering, developing, and promoting human recording artists, whose artistic contributions are the bedrock of the recorded music industry and the music we listen to today.” Both companies are accused of training their respective AI models in a way which “involved copying decades worth of the world’s most popular sound recordings and then ingesting these copies into [the] AI model[s]”.

Both developers are alleged by the plaintiffs to have been “deliberately evasive” about the data used to train their respective AI models. Suno, Inc executives previously said that its model is trained on a “mix of proprietary and public data” which is “fairly in line with what other people are doing”, whereas Uncharted Labs executives have publicly stated that their model is trained on “the best quality music that’s out there” – with the labels arguing, in Uncharted Labs’ case, that they can only be referring to the labels’ copyright-protected sound recordings.

In both lawsuits, the plaintiffs allege that it is “obvious” that copyrighted sound recordings are being used to train the models, pointing to the high-quality digital audio files produced by both models as evidence to support their claims.

Goodness, gracious…

In perhaps the most compelling evidence against the AI developers, both lawsuits outlined multiple examples of instances where prompts were plugged in to the software with a specific target copyrighted sound recording in mind, citing specific artists, subject matter and/or vocal styles, and the respective models repeatedly generated output files which closely matched that target sound recording.

In testing Udio, one such example used the prompt “my tempting 1964 girl smokey sing hitsville soul pop”, combined with lyrics by The Temptations, Udio created a song called “Sunshine Melody” which was allegedly “instantly” recognisable as resembling The Temptations’ 1965 classic, “My Girl”. Similar tests were carried out on Suno: when the original lyrics to Jerry Lee Lewis’ “Great Balls of Fire” were inputted into the platform along with the prompt “1950s rock and roll, jerry lee lewis, sun studio”, an audio file which included not just the lyrics but also the rhythm and vocal leaps from the original song was produced in the digital output, titled “You shake my nerves and you rattle my br”.

More airtime for fair use

According to the lawsuits, Suno, Inc and Uncharted Labs have both put forward the “fair use” defence in pre-action correspondence with the plaintiffs – something which, the plaintiffs argue, is a tacit admission to the allegations, as fair use is only relevant as a defence to unauthorised copying.

Anyone who is keeping tabs on the number oflawsuits making their way through courts internationally regarding the unauthorised use of copyrighted materials to train AI models, including the ongoing dispute between the New York Times and OpenAI, will be familiar with the concept of “fair use”, a creature of US law which balances a number of factors to determine whether unauthorised use of copyrighted material is “fair”. In reaching their judgment, US courts will consider in particular the purpose and character of the use, the nature of the copyrighted work, the substantiality of the portion used in relation to the copyrighted work and the effect of the use upon the market or for the value of the copyrighted work.

How might this play out in the UK?

The UK’s Copyright, Designs and Patents Act 1988 sets out a very narrow group of exceptions to copyright protection, including fair dealing, which permits the copying of copyright-protected material for non-commercial research and private study, criticism or review, or news reporting. The key question to be asked is how a fair-minded and honest person would have dealt with the copyright-protected work.

Relevant considerations, if this case were to be brought in the UK, would relate to how the use of the major labels’ sound recordings to train Suno and Udio would impact the market for human-made music, and whether the proportion of those sound recordings used to do so was reasonable and appropriate. A court would have to take a view on labels’ submissions that the technologies “threaten enduring and irreparable harm to recording artists, record labels, and the music industry”, and would likely demand greater scrutiny of the training models to assess whether the use of the sound recordings to do so – if proven – was reasonable and appropriate.

While the UK courts have not yet had the opportunity to consider the legality of using copyright-protected musical works to train AI models to date, the question of the relationship between copyright and generative AI remains a live one. With the Labour Government promising to strengthen AI regulation in its 17 July King’s Speech, perhaps an enhanced regime will dare to clear up the waters – but only time will tell.