Several US music industry organisations have filed a court brief in the legal battle between book authors, newspaper owners and OpenAI, arguing that training artificial intelligence (AI) models on copyrighted works should never count as fair use under American copyright law. The outcome matters to the wider music business because the same question underpins its own lawsuits against AI companies.
Who filed the brief and why#
The brief was filed by the following organisations:
- RIAA (Recording Industry Association of America)
- NMPA (National Music Publishers’ Association)
- A2IM (Association for Independent Music)
- SONA (Songwriters of North America)
- ARA (Artist Rights Alliance)
The filing is an amicus brief, submitted to the court overseeing the case to set out the music industry’s current arguments. If AI training were considered fair use, AI companies could use existing works to train their models without permission from any creators or rightsholders.
The case centres on OpenAI’s copying of books and journalism to train its AI models. Nearly all of the copyright lawsuits filed against AI companies in the US turn on the fair use argument, including the music industry’s litigation against Suno, Udio and Anthropic.
Government position in the case#
The Department of Justice under President Donald Trump has previously told the judge that a ruling finding OpenAI’s training was not fair use would be “inconsistent with basic copyright law principles” and would “severely hamper” the “progress of science and arts”.
Because the Department of Justice has sided with OpenAI, the music organisations are backing the authors and newspaper publishers. They do not criticise the administration directly, but the brief rejects its position. It argues that the purpose of copyright law is “to encourage humans to produce creative works for the public’s benefit”.
The filing says that allowing AI companies to copy existing works without permission in order to “generate digital imitations” does the opposite, and “affirmatively undermines the constitutional incentives for humans to author creative works”.
The brief opens by comparing the OpenAI case with lawsuits filed by record labels, music publishers and musicians against music AI companies. It notes that “like defendants in this case, the services targeted by the music community insist that training on their copyrighted works is fair use”.
On the AI companies’ fair use arguments, the filing states “they are wrong”. Because members of the organisations “own some of the most cherished copyrights in modern culture”, the brief sets out what it calls “defects” in OpenAI’s fair use defence.
Some of the key arguments against fair use were already put forward in the Suno, Udio and Anthropic cases. The organisations say these arguments are being refined and reprioritised as the wider debate over AI companies’ copyright obligations evolves.
Transformative use#
When weighing a fair use defence, US courts consider whether the use of a copyrighted work is “transformative”. The question is whether a new work created from an existing one is sufficiently different from the original. The more transformative a use, the more likely it is to be fair use.
AI companies have repeatedly argued that their models’ outputs are not exactly the same as any content used in training, and that this makes the use clearly transformative.
The music organisations say OpenAI and other AI companies are applying this test incorrectly. Citing the Supreme Court’s 2023 fair use ruling involving the Andy Warhol Foundation, they say a transformative work must “serve a purpose akin to that of the original work”.
The brief also argues that courts must examine the purpose of OpenAI copying any individual book or article, meaning the reason the company needs to make that copy. It states:
“An AI model does not ingest romance poetry or bluegrass to learn about language or music at large, but rather to generate substitutes that read and sound like them in practice, thus achieving the same, non-transformative purpose.”
Market dilution#
Market dilution considers whether a new work created by copying an existing work is likely to harm the value of the original. US courts must weigh this as one of the fair use factors, and it is probably the factor that most favours copyright owners in the AI debate.
In an earlier case involving book authors and Meta, the judge said market dilution arguments could defeat an AI fair use defence. If a use of a protected work can be shown to harm that work’s market value, it is less likely to be fair use.
The music industry argues that when AI companies copy existing music to train generative AI models, which then output new music, the market value of the existing music is negatively affected.
