A text chatbot can face a song-copyright claim without even singing a note. The Elvis Presley Anthropic lawsuit, filed on October 23, 2026, brings alleged copying by Claude’s developer before a federal court. Companies and trusts that say they control rights to 274 compositions recorded by Presley accuse Anthropic of copying lyrics to develop Claude, retaining unauthorized copies, and producing infringing text. The claims concern the songs’ underlying music and lyrics. They do not accuse Claude of cloning Presley’s voice.
For AI developers, the dispute reaches from data collection to what a chatbot returns to users. For people using Claude to write, it raises a different concern: whether generated lyrics reproduce someone else’s work. A recognizable lyric in a response does not, by itself, establish where the model encountered it, what copies Anthropic kept, or whether a particular use was legally permissible.
What the Elvis Presley Anthropic lawsuit alleges
The seven plaintiffs include ABG EPE IP LLC, music publishers, trusts, and an individual who say they own or control rights in the compositions. They sued Anthropic in the U.S. District Court for the Northern District of California, case 5:26-cv-12473. The complaint filed on October 23 asserts three claims: direct copyright infringement, circumvention of technological protections, and removal or alteration of copyright-management information.
The affected catalog includes songs such as Can’t Help Falling in Love and Don’t Be Cruel. These are works Presley recorded, not necessarily songs he wrote. Under U.S. copyright law, a song’s composition and lyrics are legally separate from the particular sound recording of an artist performing it. The two rights are commonly owned and licensed separately. A text chatbot can potentially reproduce protected lyrics without producing a second of audio.
The plaintiffs seek statutory damages of up to $41.1 million for their principal infringement claim, with additional sums sought under the Digital Millennium Copyright Act. They also ask for an injunction, an accounting of relevant training materials, and the destruction or impoundment of copies they characterize as unauthorized, potentially including material in datasets and model infrastructure. Those are requests for relief, not orders already issued. Anthropic disputes the claims and intends to defend the case fully, according to its response to Pitchfork.
Training data, stored copies, and outputs are three different questions
▪ Did Anthropic copy the songs to train Claude?
The plaintiffs allege that lyrics reached Anthropic through large collections of online text, including Common Crawl and The Pile, and through other sources. They point to a YouTube-subtitles dataset that reportedly contains some works on their list. They also invoke evidence about Anthropic’s use of pirated book collections from a separate lawsuit.
Showing that a dataset contained a song is one step. Establishing that Anthropic used that text in a particular Claude training run requires evidence connecting the two. The complaint relies in part on allegations made on information and belief. Its list of 274 compositions establishes the scope of the copyrights being asserted, not 274 individually demonstrated instances of training-data inclusion. The filing names particular songs and datasets but relies in part on allegations still to be proved.
Training logs, dataset snapshots, acquisition records, and preprocessing records could connect a disputed lyric to the material actually used. Even if the plaintiffs establish copying, the court would still need to consider whether the particular use qualifies as fair use.
▪ Did it keep copies after training?
The complaint says Anthropic maintained a general-purpose internal library containing the disputed works, separate from any single training run. Keeping a copy indefinitely can receive a different legal assessment from copying material for a defined training process. It also alleges that each new model contains unauthorized copies, a proposition requiring both technical evidence and legal analysis. Model weights cannot simply be treated as ordinary text files containing a library of lyrics.
Bartz v. Anthropic, decided in June 2025 offers a relevant precedent. In that book-copyright case, a federal judge held that Anthropic’s specific LLM-training use was fair use, while rejecting fair use for acquiring and indefinitely keeping pirated books in a central library. The ruling did not establish that every AI training use is lawful, or that the Elvis lyrics were among those books. The sources and uses of copies must be examined separately.
▪ Did Claude reproduce protected lyrics?
The plaintiffs also claim Claude has generated recognizable lyrics and derivative versions of songs, including after Anthropic introduced output restrictions. They describe a prompt asking Claude to recreate songs “for a modern audience” and cite the chatbot’s own assessment that its rewrites resembled originals. The filing, however, does not publicly provide a complete, song-by-song set of Elvis outputs demonstrating infringement for all 274 compositions.
A chatbot’s assessment of its own answer is no substitute for a copyright analysis. Its response also cannot establish its entire training history. A lyric might have appeared in training material, in a user’s prompt, in retrieved context, or in several overlapping online sources. Evaluating the claim would require the precise prompts and outputs, the model version, and the relevant records. Without those details, readers cannot tell what the system was given or how much of the response came from the disputed work.
If Claude reproduces substantial protected expression, there may be an output-infringement case regardless of whether it can sing. Yet writing in a familiar musical style is not automatically infringement. The contested question is whether the generated text copies protected expression closely enough, and what responsibility Anthropic bears for that output.
The DMCA claims target access controls and missing credits
The lawsuit also invokes provisions of the Digital Millennium Copyright Act that reach beyond straightforward copying.
Under 17 U.S.C. § 1201, the plaintiffs allege Anthropic circumvented measures protecting access to copyrighted material. The filing mentions website access restrictions, including authentication systems, paywalls, and anti-bot controls. Proving that a company scraped a website is not necessarily enough. This claim requires an applicable technological protection measure and evidence of legally relevant circumvention. A site’s terms of service or robots.txt instructions should not automatically be treated as an access-control system meeting the statute’s requirements.
Under 17 U.S.C. § 1202, the plaintiffs allege that Anthropic stripped identifying details such as song titles, composer credits, and copyright ownership information when preparing data. The law addresses intentional removal or alteration of qualifying copyright-management information, along with a knowledge requirement tied to facilitating or concealing infringement. Losing page metadata through an automated cleanup process is not, by itself, proof that every element of this claim is satisfied.
Both claims put Anthropic’s data-acquisition tools and cleanup practices under scrutiny. They also offer potential civil remedies separate from ordinary copyright damages, if violations are proved.
Can Anthropic argue that training on lyrics is fair use?
Yes. The Copyright Act’s four-factor test considers the purpose of the use, the nature of the work, how much was copied, and the effect on relevant markets. There is no automatic answer covering every AI training use. The court must evaluate the particular copying and the evidence supporting each factor.
The Elvis plaintiffs can argue that song lyrics are highly creative, that entire works were copied, and that unlicensed training and generated substitutes threaten markets they could otherwise license. Anthropic can argue that training models to learn language patterns is transformative and that the plaintiffs must substantiate the claimed market harm. How Anthropic acquired the lyrics, which copies it retained, and what Claude generated will shape those arguments. A defense for one use would not automatically resolve every other alleged act of copying.
The dispute is already playing out in related proceedings. In an October 21, 2026 hearing involving music publishers, Anthropic and opposing publishers argued over transformative training and whether generated lyrics harm the market. That litigation has not resolved the Presley plaintiffs’ claims.
What the Elvis lawsuit means for developers and Claude users
As of the time of publication, this is a newly filed civil lawsuit, not a finding that Claude infringed all 274 works. It does not itself prohibit users from asking Claude about Elvis or require a new consumer license.
For model developers, the practical lesson is to preserve evidence about data provenance. Which sources were downloaded, under what access conditions, what identifying information was removed, and which copies remained after training? These details can determine the strength of a defense long after a model has been released.
For creators and publishers, generated lyrics should be checked before commercial use. A model’s refusal to quote an entire copyrighted song is no guarantee that a rewrite avoids protected expression. Conversely, a model producing a recognizable line cannot establish on its own that a particular copyrighted file was used in training.
The Elvis copyright fight will turn on copies and evidence
The court may eventually have to decide whether particular copying was unauthorized, whether fair use applies, whether access controls or credits were improperly handled, and whether particular outputs infringe. Until that evidence is tested, declaring either side the winner would be premature.
The records to watch are the ones connecting particular songs to Anthropic’s datasets, retained files, and Claude’s outputs. Those records could establish what was copied and why. Presley’s voice makes the story recognizable, but the court’s copyright analysis will turn on the text behind the songs and what Anthropic did with it.
Explore more from Popular AI:
Start here | Local AI | Builds & gear | Autonomy & policy | Fixes & guides | Popular AI podcast