When AI Learns the Lyrics: A Copyright Fight Every Author Should Watch

When AI Learns the Lyrics: A Copyright Fight Every Author Should Watch

A courtroom argument about song lyrics might sound a long way from the everyday business of writing a novel. I don't think it is. On 21 October, Reuters reported that major music publishers and Anthropic were preparing to argue over whether using copyrighted lyrics to train an AI model qualifies as fair use under US law. The immediate dispute concerns music, but the underlying question travels straight into publishing: who gets to decide how creative work is used to build commercial AI systems?

The question behind the courtroom door

According to Reuters, publishers including Universal Music Group-related interests, Concord and ABKCO are challenging Anthropic over copyrighted lyrics, while Anthropic argues that training is transformative and supports innovation. These are competing legal positions, not a settled finding that either side has won this particular dispute. That distinction matters. We are watching courts work through different factual circumstances, different collections of material and different kinds of alleged harm. There is no single judgment that neatly answers every question about books, lyrics, images and AI.

For authors, the instinctive reaction may be to ask whether a machine has copied a sentence. But the bigger argument is about what happens before a reader ever sees an answer: a system can acquire value by learning from vast quantities of human-created material. The creators of that material may never have agreed to the use, negotiated terms or seen any payment. Developers, meanwhile, argue that learning statistical patterns is not the same as selling a copy of the original work. That is the fault line.

Why books and lyrics are not identical

We should resist pretending that a lyric and a 100,000-word novel raise precisely the same issues. Lyrics are short, memorable and often quoted verbatim. Novels involve plot, voice, character, structure and an enormous amount of accumulated craft. A chatbot reproducing recognisable song lyrics presents a different practical question from a model generating a generic fantasy scene. Yet both can raise questions about the provenance of training data, permission, substitution and the market for licensed material.

The distinction also matters between training and output. One argument asks whether protected works were lawfully obtained and used during training. Another asks whether a particular output reproduces protected expression or competes unfairly with the original. Conflating those questions makes for a fiery social post but rarely helps authors understand what is actually being decided.

The publishing industry is already moving

This is not an abstract future debate. UK Publishers' Licensing Services reported in June that more than 250 publishers had opted into its collective AI licensing scheme. In April, the Authors Guild published updated model contract clauses addressing AI training, subsidiary rights, translations, narration and the handling of manuscripts by publishers. These initiatives take different approaches, but they recognise a common point: rights and permissions cannot be left as an afterthought.

Licensing is potentially useful, but a deal announced by a publisher is not automatically a deal that answers every author's concerns. Who controls the rights? Which works are included? Is the payment meaningful? Can an author decline? Does an existing publishing contract even grant the rights someone proposes to license? The answers depend on the agreement. Authors should not assume that a headline about an industry deal settles those questions on their behalf.

Using AI as an author is a separate choice

I write science fiction and I am interested in technology. I use AI to explore ideas, organise research and support the business of being an author. I also believe the story, the characters, the choices and the actual creative direction must remain mine. Those two positions are not contradictory. The use of a tool in a writer's workflow is different from giving a third party unrestricted permission to absorb that writer's books into a commercial training dataset.

It is also different from presenting machine-generated fiction as entirely human-written. If AI contributes substantive prose, artwork or narration, readers, collaborators and contractual partners may reasonably expect disclosure and appropriate rights checks. If it helps with administration, research discovery or alternative outlines that the author independently develops, the conversation is more nuanced. We need a vocabulary that recognises these differences instead of dividing everyone into two hostile camps.

Four practical questions worth asking

First, look at your contracts. Find the clauses that deal with electronic rights, subsidiary uses, data mining, translation and future technologies. If a clause is unclear, seek qualified advice rather than guessing. Second, keep a record of your creative process: drafts, notes and major revisions. This is useful for ordinary authorship disputes as well as questions involving AI.

Third, ask any publisher, agent or service provider what happens when your manuscript is uploaded to an AI tool. Is it retained? Can it be used for training? Who can access it? Fourth, decide your own boundary for AI assistance and explain it plainly. A consistent policy is more useful than trying to improvise an answer each time somebody asks.

The principle I would keep

The legal result of this particular music case is still to come. It would be premature to call it a decisive victory for either creators or developers. But its importance to authors is already clear. Innovation should be possible without treating creative work as a resource that nobody owns, and copyright protection should be discussed with enough precision to distinguish genuinely new uses from unauthorised copying.

For me, the sensible destination is neither a blanket rejection of AI nor a free-for-all. It is an environment where authors can choose how their work is used, where licensing terms are intelligible, and where readers know what they are buying. The technology is moving quickly. Our standards for consent and accountability need to keep up.

Sources and further reading

Reuters, 21 October 2026 — Anthropic, music labels to make key arguments in AI training case: https://www.reuters.com/legal/legalindustry/anthropic-music-labels-make-key-arguments-ai-training-case-2026-10-21/

The Authors Guild, 29 April 2026 — New AI model clauses: https://authorsguild.org/news/authors-guild-releases-new-ai-model-clauses-and-issues-updates/

Publishers' Licensing Services — AI licensing updates: https://www.pls.org.uk/news-events-policy/news

About the author: https://www.121minutes.com/about

For more AI & The Author commentary, subscribe: https://www.121minutes.com/subscribe

AI & Editorial Transparency

AI & The Author is edited and published by Rob Frankson. Artificial intelligence is used to assist with news research, initial drafting, content organisation and supporting imagery. All articles are reviewed and, where necessary, edited by Rob Frankson before publication. The opinions, editorial position and final decision to publish remain the author's.