Eight music industry organisations in the US have teamed up to formally support the music publishers who are suing AI company Anthropic. They have submitted a so called ‘amicus brief’ with the court setting out why training an AI model that generates music with existing songs and recordings is not and can never be ‘fair use’ under American copyright law, because of market dilution.
Anthropic has been sued by the Universal, Concord and ABKCO music publishing companies for copying existing lyrics when training its Claude AI model without securing permission from the music industry. However, if AI training is fair use, as Anthropic claims, then it can copy as many lyrics as it likes when training Claude without getting permission from any songwriters or music publishers.
There are various factors to consider when assessing if any one use of a copyright protected work is fair use. But “undoubtedly the single most important element”, the music industry organisations say in their new filing, “concerns the effect that unlicensed copying of protected works will have on the market or ‘the potential market for or value of’ those works”.
And the unlicensed copying of lyrics - or indeed sheet music or sound recordings - to develop generative AI models that then generate new music “has enormously damaging effects on the music industry”, the organisations continue. Because “it floods the market with songs that substitute for and divert royalties away from the very works copied by AI developers to power their products”.
Universal, Concord and ABKCO recently asked the judge overseeing their lawsuit against Anthropic to reject the AI company’s fair use arguments and rule in their favour.
In their own filing opposing the fair use defence, the three publishers cited the ruling in a legal battle between a group of authors and Meta over its training of generative AI models. In that case Judge Vincent Chhabri actually accepted Meta’s arguments that its AI training was fair use, but said that a fair use defence could be defeated if a decent market dilution argument could be presented.
Crucially, Chhabri said such an argument could involve a wider interpretation of ‘market dilution’. So rather than a copyright owner having to demonstrate how the unlicensed copying of a single work had negatively impacted on the market value of that specific work, instead the negative impact could be on a category of works that the original content belonged to. Which, in this case, would be lyrics in general.
The three publishers insist that that is what is happening when Claude is trained on existing lyrics in order to generate new lyrics. And that claim is strongly supported in the new amicus briefing filing by the Recording Industry Association Of America, National Music Publishers’ Association, American Association Of Independent Music, SoundExchange, Songwriters Of North America, Black Music Action Coalition, Music Artists Coalition and Artist Rights Alliance.
AI-generated songs are “one-to-one market substitutes for copyright-protected songs”, the industry groups argue. Because, for example, with streaming the digital platforms “calculate royalty ‘pools’ which are then allocated to copyright owners on the basis of a copyright owner’s percentage share of total ‘plays’ of songs on the service”.
And, “if the percentage share of total plays of a copyright owner’s songs decreases due to plays of AI-generated songs that cannot be identified as AI-generated, the human creator receives less money”. And “the financial impact” of market substitution of this kind “is expected to be significant”.
The eight trade groups also present a second argument for market dilution. They say that there is an emerging market for licensing music to AI companies for training generative AI models, and if companies like Anthropic ignore that licensing market, that dilutes its value.
Anticipating Anthropics’s response to that argument, they say, AI companies “have argued that no market for licensing copyrighted works as 'training data' exists or is likely to develop, and claim that it is difficult, if not impossible, to identify and license works from copyright owners at scale”.
However, they add, any complexities in music licensing are irrelevant because there is “no such thing as fair use by reason of necessity”. Plus “this is not an accurate picture of licensing, at least in the music industry”, because “major players in the music marketplace have already announced several groundbreaking agreements to license their copyrighted works to ‘train’ music-focused AI systems”.
We await to see how the judge responds to the arguments put forward by the three music publishers and their supporters from across the music industry.
And also how many of Anthropic’s friends in the tech sector seek to also intervene in the case and present arguments in favour of the fair use defence. Of most interest is the position taken by those in the tech sector who are also involved in music streaming, such as Apple, Amazon and Google. The trade body for music streaming services in the US - DiMA - is not part of the music industry’s amicus brief. Will the big tech owners of many of those services seek to intervene on the other side of the debate?
But if the fair use arguments presented by the publishers and their music industry allies do prove successful, they will provide a template for pretty much every lawsuit filed by the music business against AI companies that use unlicensed music in their training in the US.
Especially as President Donald Trump has said it is for the courts to settle the debate around AI training and fair use, despite the White House being generally more sympathetic to the AI companies on all this.
Of course, if the judge was to rule against Anthropic, but the AI company appealed, it could take years to get a final ruling on this matter.
However, in the short term, any legal uncertainty for AI companies is beneficial for the music industry, given the potential for mega-damages - of billions or even trillions of dollars - under US copyright law. After all such uncertainty has already helped pressure AI companies that previously said they didn’t need music licences into licensing negotiations.