Anthropic Elvis Presley Lawsuit Targets Claude Training on 274 Songs

The King's catalog goes to court

The companies that own Elvis Presley's songs have sued Anthropic. Their federal lawsuit says the company behind the Claude chatbot copied the lyrics of 274 songs Presley recorded, used them to train its models without a license or payment, and built a system that still produces infringing versions of them.12 The complaint was filed on October 23 in the U.S. District Court for the Northern District of California. The lead plaintiff is ABG EPE IP LLC, a unit of Authentic Brands Group, the licensing company that controls Presley's name and likeness and part of the publishing rights to his recordings.2 Raleigh Music Publishing, Abilene Music, members of the Aberbach family and trusts linked to the Aberbach family and songwriter Luigi Creatore joined the suit.10

The list of songs includes "Can't Help Falling in Love," "Don't Be Cruel," "All Shook Up," "A Little Less Conversation" and "Return to Sender."114 Anthropic gave every outlet the same reply: it disagrees with the claims and will fight the case in full.71213

The case adds a famous name to a fight the music industry was already having. Its real weight, though, comes from how it was built. It reuses findings from earlier Anthropic cases, adds new claims under the Digital Millennium Copyright Act, and asks for a remedy that would reach into the trained models themselves.

What the complaint alleges

The suit makes three main claims. The first is direct copyright infringement. The other two fall under the DMCA. One says Anthropic got around technical protections on lyrics sites. The other says it removed copyright management information, meaning song titles, songwriter and publisher credits, and copyright notices, before using the text for training.210

The plaintiffs say Anthropic gathered text "at an industrial scale" from lyrics sites and from public datasets such as Common Crawl and The Pile. They specifically claim that the YouTube-subtitle portion of The Pile contains Presley lyrics.2 The San Francisco Chronicle adds that the filing names the Internet Archive as another source. It also reports the claim that copying happened at every stage: when material was collected, cleaned, stored and fed into each new Claude model.4

The credit-stripping claim is the most detailed part of the filing. According to the complaint, Anthropic tested at least three text-extraction tools. It picked one called Newspaper because it removed copyright information so thoroughly, and passed over another, jusText, because it kept footnotes with copyright notices, which Anthropic allegedly called "boilerplate" and "useless junk."2 The plaintiffs argue that engineers knew, or should have known, that training without this information would stop Claude from reproducing it, making infringement harder to spot.2

The filing also leans on Bartz v. Anthropic, the authors' case. It cites a July 2025 ruling in that case to allege that Anthropic used BitTorrent to download pirated books from Library Genesis, some of which allegedly contain the lyrics. Because BitTorrent shares pieces of files with other users during a download, the plaintiffs also accuse Anthropic of distributing the songs.2 The complaint further describes an internal library that the plaintiffs call a permanent stockpile Anthropic can reuse for future training.23

The damages figure depends on the outlet

Coverage reports two different headline numbers, and both are correct. The New York Times, Pitchfork, People and the BBC lead with about $41 million.121357 Variety and Music Business Worldwide report $48.6 million.112 The $41.1 million is the maximum statutory award for willful infringement, $150,000 per song across 274 songs. The two DMCA claims add stated totals of $685,000 and $6.85 million, which brings the overall figure to just over $48.6 million.210 These are amounts the plaintiffs are asking for, not anything a court has awarded.10

Outlets also differ slightly on where the case was filed. The Los Angeles Times says federal court in San Francisco, while the Chronicle says San Jose. Both courthouses belong to the same Northern District.34

The money matters less than the injunction the plaintiffs want. They ask the court to order unauthorized copies impounded or destroyed, "including copies retained in training datasets, internal repositories, model weights, and server infrastructure."2 Model weights are the learned values that make up a trained model, so granting that request in full would reach Claude itself.2 No court has said it would go that far, and it would be surprising if one did. Still, putting weights in the request raises the stakes of any settlement talks.

The weak point: no Elvis output

Most coverage repeated the complaint's most quotable line. Asked to recreate songs "for a modern audience," Claude allegedly said its versions "keep landing very close to the originals," with the same structure and hooks and only light rewording.713 Music Business Worldwide was the outlet that checked where that line came from, and its finding changes how the case looks. The exchange did not involve a Presley song. The complaint takes it from Round Hill Music's suit against Anthropic, filed by the same legal team in August, where the songs in question included the Goo Goo Dolls' "Iris."2 The Presley filing gives no example of Claude reproducing or rewriting any of the 274 songs. Instead it alleges, "on information and belief," that Claude has produced infringing output for all of them, and says the evidence is in Anthropic's own records.210

Several outlets presented the quote as if it were evidence about Elvis.68 It is not, and readers should treat it as evidence of a pattern. The output side of this case is the weaker side unless discovery turns up Presley-specific examples. The training-side claims and the credit-stripping claims are stronger, because they draw on records already developed in other cases.

The plaintiffs also go after Anthropic's guardrails. They say the filters added after music publishers first sued in October 2023 cover only a small set of well-known songs and can be bypassed by rewording a request. They add that Claude Haiku 4.5 and Sonnet 5 refuse to print lyrics but will readily write derivative versions.2 Under copyright law, a derivative work can infringe as surely as a verbatim copy.

A litigation strategy, not a one-off

The lead lawyer explains a lot about this case. Richard Busch of Adams & Reese won the "Blurred Lines" verdict for Marvin Gaye's family.2 He sued Anthropic for Round Hill in August, a case Variety puts at $1 billion.11 His team filed for Reservoir on October 21, two days before the Presley suit.27 This looks like a deliberate plan: use one detailed body of factual claims and apply it catalog by catalog, so each publisher can collect statutory damages on its own works without waiting for the larger industry cases.

Those larger cases keep growing. Universal Music Publishing Group, Concord and ABKCO sued in October 2023 over about 500 songs, then filed a second suit in January 2026 covering more than 20,000 works.2 Mashable reports that Sony Music Publishing and Warner Chappell sued in August. The Los Angeles Times notes that the Presley filing counts at least six pending cases in San Francisco federal court over music used for AI training.3 Pitchfork describes this as one of the first high-profile suits from the owner of a single artist's catalog.13 That is a real change, because it shows the legal theories have become simple enough for individual rights owners to bring.

Fair use, settlement and timing

The central legal question is still open. AI developers, Anthropic included, argue that training on copyrighted material is fair use. Courts have not settled that question across the dozens of pending suits.14 What has shifted is how much trouble Anthropic's past data collection can cause. The company agreed to pay $1.5 billion to settle the authors' class action over pirated books, a deal most outlets date to last year.11116 (Mashable dates it to July, which conflicts with the others.) Another authors' case, Shakespeare v. Anthropic, filed in June, makes similar claims about pirated books.8 By putting how the data was obtained at the center, rather than the act of training, the Presley plaintiffs are using the argument that has already cost Anthropic the most.

The timing adds pressure. Reports differ on the order of events: the Chronicle says the suit came a day after Anthropic filed paperwork for a proposed IPO, while CBC says the IPO filing came the following Monday.4 Either way, the two happened within days of each other. Variety reports that the complaint openly contrasts Anthropic's public talk about AI's value to society with an offering that could be worth up to $2 trillion.11 As a company heads toward a public listing, every unresolved catalog claim becomes a risk it has to disclose, and each one strengthens the plaintiffs' hand in settlement talks.

Why it matters beyond music

The policy lesson here is about record-keeping more than lyrics. The most damaging claims in this complaint concern how the data was handled: which extraction tool was chosen, why copyright notices were treated as junk, and what happened to a permanent store of collected text. If courts give real weight to the credit-stripping claims, removing copyright information during training becomes a liability of its own, separate from whether training is fair use. That would push developers toward licensed sources and documented data pipelines, a change copyright policy debates have not yet produced. As of now, the plaintiffs have strong evidence about how Anthropic obtained the data and thin evidence about what Claude outputs for these songs, and how the case turns out will probably depend on which of those a judge treats as the core issue.