Sony Music Publishing and Warner Chappell Music have sued Anthropic, escalating the music industry’s fight over copyrighted material allegedly used to train generative AI systems.
According to Reuters, the publishers filed the case in California federal court, accusing Anthropic of using copyrighted song compositions to train Claude. The complaint alleges that Anthropic obtained lyrics and sheet music through torrent downloads and used works associated with artists including The Beatles, Taylor Swift and Michael Jackson.
The Guardian reported that the lawsuit concerns “tens of thousands” of copyrighted works, including songs such as Mariah Carey’s All I Want for Christmas Is You, Ain’t No Mountain High Enough and Survivor’s Eye of the Tiger.
Publishers challenge Anthropic’s training methods
The dispute goes beyond whether Claude can reproduce copyrighted lyrics in its outputs.
Reuters reported that the publishers allege Anthropic used pirated material for training and that Claude can reproduce some copyrighted lyrics “verbatim” when prompted.
The complaint also alleges that copyrighted material helped Claude generate lyrics that could compete with legitimate works, raising questions not only about model outputs but also about the datasets used to develop generative AI systems.
The Guardian said the lawsuit names Anthropic CEO Dario Amodei and co-founder Benjamin Mann alongside the company and alleges the defendants used methods including “torrenting, scraping and downloading” copyrighted material.
The financial exposure could be substantial. Reuters reported that the publishers are seeking up to $150,000 for each infringed copyright, as well as an order preventing Anthropic from continuing to use the works at issue.
The Guardian reported that possible damages could reach billions of dollars, with the complaint also seeking $25,000 for each alleged removal or alteration of identifying information associated with protected works.
Anthropic pushes back
Anthropic has rejected the claims.
A company spokesperson shared that the case was the third lawsuit from the same lawyers and repeated allegations already before courts. Anthropic said it would defend itself “robustly” and maintained that training AI systems on copyrighted material constitutes fair use.
The dispute follows Anthropic’s $1.5 billion settlement with authors last year, which Reuters described as the first settlement by an AI company in the growing wave of copyright litigation surrounding model training.
The new complaint refers to Anthropic’s reported $2 trillion valuation while arguing that the earlier settlement failed to deter the alleged infringement.
Training data is becoming an infrastructure risk
For AI builders, the lawsuit highlights an issue moving rapidly from legal departments into core model-development strategy: how training data is sourced, acquired and governed.
Authors, publishers, music companies and news organizations have already brought a growing wave of copyright cases against technology companies over protected works used in AI training. Universal Music Group has pursued separate litigation against Anthropic involving song lyrics.
For founders and engineering leaders, the larger risk is no longer limited to whether a model reproduces copyrighted material in an answer. Courts are increasingly examining the upstream datasets and acquisition methods used to build the models themselves.
That makes data provenance, licensing and dataset governance part of the technical and financial architecture of generative AI—and potentially one of its most expensive unresolved risks.
