TL;DR
Anthropic has been sued by music publishers who allege the AI company used copyrighted lyrics from tens of thousands of songs without authorization. The case is part of a broader wave of copyright litigation against AI developers over training data.
Anthropic, the company behind the Claude AI chatbot, is being sued over allegations that it used copyrighted song lyrics from “tens of thousands” of songs without permission, according to a report by The Guardian. The lawsuit, brought by music publishers, escalates the legal pressure on AI developers over the data used to build and train their models.
The lawsuit alleges that Anthropic reproduced and made use of lyrics owned by music publishers without licensing them, with the plaintiffs putting the scale of the alleged copying in the “tens of thousands” of works. The word “theft” appears in the reporting on the case as a characterization of the publishers’ claims — it is an allegation, not an established finding.
The core of the dispute mirrors other copyright cases now working through US courts: publishers argue that Anthropic’s models were trained on, and can output, copyrighted lyrics, and that this constitutes infringement. Anthropic and other AI companies have generally argued that using publicly available text for training constitutes fair use under US copyright law, a defense that remains untested at final resolution in most of these cases.
What is confirmed at this stage: a lawsuit has been filed, and the publishers have made specific allegations about the scale and nature of the lyrics use. What is not confirmed: whether any infringement occurred. That is a question for the court, and Anthropic has not conceded the publishers’ claims.
Why the Lyrics Case Matters
The case matters for two reasons. First, song lyrics are among the most precisely copyrighted forms of text — short, heavily licensed, and aggressively enforced by rights holders. If courts find that training on or reproducing lyrics constitutes infringement, AI companies face damages models that scale with the number of works involved, which is exactly why the “tens of thousands” figure is central to the plaintiffs’ framing.
Second, the outcome could shape the still-unsettled question of whether AI training on scraped text counts as fair use. Dozens of suits from authors, news organizations, visual artists, and music publishers are pending against OpenAI, Meta, Google, and others. A ruling against Anthropic in a lyrics case would strengthen the bargaining position of all rights holders seeking licensing fees — something major publishers have already begun negotiating directly with AI firms.
AI training data copyright protection
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Broader Copyright Litigation Wave
This lawsuit is one thread in a larger wave of copyright litigation against AI developers that began after ChatGPT’s release in late 2022. Plaintiffs have included book authors, news outlets, stock photo companies, and record labels, all arguing that AI models were built on their work without consent or compensation.
Music rights holders have been among the most litigious. Labels and publishers have sued AI companies over voice cloning, generated music, and lyrics, and have pushed for licensing deals as an alternative to courtroom fights. Anthropic, founded as a safety-focused AI lab and backed by Google and Amazon, now faces the same fundamental legal question as its larger competitors: whether the data scraped to train large models was used lawfully.
Early procedural rulings in related AI copyright cases have sometimes required defendants to disclose training data details, but no US court has yet delivered a final ruling on the fair-use question at the heart of these disputes.
“Anthropic sued over alleged theft of ‘tens of thousands’ of songs”
— The Guardian (reporting on the lawsuit)
As an affiliate, we earn on qualifying purchases.
What the Lawsuit Does Not Yet Establish
The word “theft” in the reporting reflects the plaintiffs’ characterization. No court has found that Anthropic committed infringement, and Anthropic disputes the claims through its defense. Several key questions remain open: exactly which works are alleged to have been copied, how the lyrics entered Anthropic’s training corpus, whether the models reproduce lyrics verbatim in output, and whether any use qualifies as fair use.
It is also unclear whether the case will reach a full trial or be resolved through settlement, as some other AI copyright disputes have been. Damages figures reported at the lawsuit stage are typically what plaintiffs request or estimate, not amounts awarded.
As an affiliate, we earn on qualifying purchases.
The Legal Road Ahead
The case will now move through procedural stages: Anthropic’s response to the complaint, potential motions to dismiss, and — if the case survives — discovery, in which the publishers would seek access to Anthropic’s training data and internal records. That discovery phase has proven contentious in parallel AI copyright cases, as AI companies argue training datasets are trade secrets.
Readers should watch for two milestones: any ruling on motions to dismiss or summary judgment, which could narrow or end the case, and any parallel licensing deals between music publishers and AI companies, which could reshape the litigation’s practical stakes before a verdict arrives.
As an affiliate, we earn on qualifying purchases.
Where I land
My read: this lawsuit is another strong signal that the era of building AI models on scraped data without licenses is closing, at least for well-defined, aggressively enforced content like song lyrics. Lyrics are short, easily verified against training outputs, and backed by sophisticated rights holders — a bad combination for any defendant arguing fair use. I expect this case to end in settlement or a licensing framework rather than a landmark verdict.
The strongest counterargument is Anthropic’s: training a model on text is not the same as publishing it, and US fair-use doctrine has historically protected transformative uses. If courts accept that training is transformative and outputs rarely reproduce lyrics verbatim, the publishers’ damages theory weakens considerably.
What would change my assessment: a clear court finding that Anthropic’s models reproduce substantial portions of specific lyrics on demand, or internal records showing deliberate ingestion of licensed lyric databases. Either would make this look much less like a gray-area fair-use fight and much more like straightforward infringement.
Source: Anthropic
Key Questions
Who is suing Anthropic?
Music publishers holding rights to the songs at issue, according to reporting by The Guardian. The suit alleges unauthorized use of lyrics from “tens of thousands” of songs.
Has Anthropic been found guilty of copyright infringement?
No. The case involves allegations only. Anthropic has not been found liable for anything, and the fair-use defense underlying most AI training disputes remains unresolved in US courts.
What is Anthropic’s likely defense?
AI companies in comparable cases have argued that training on publicly available text is fair use under US copyright law. Anthropic’s specific legal strategy in this case has not been fully laid out publicly.
Could this affect how Claude and other chatbots handle lyrics?
Possibly. Many chatbots already refuse to reproduce full copyrighted lyrics. An adverse ruling or settlement could push AI companies toward licensing agreements that restrict or formally permit lyric-related outputs.
How does this fit into other AI copyright lawsuits?
It is part of a broader wave of litigation from authors, publishers, artists, and labels against OpenAI, Meta, Google, Anthropic, and others, all testing whether AI training on copyrighted material is lawful.
Source: Anthropic
