Two hundred and seventy-four songs. One hundred and fifty thousand dollars each, the maximum a court can award for a willful copyright violation in the United States. The arithmetic comes to $41.1 million, and it is the first number a reader meets in the 28-page complaint filed on Oct. 23 in the Northern District of California.
The plaintiffs are a consortium rather than a single estate. Authentic Brands Group, the licensing company that bought Presley’s intellectual property and trademark rights in 2013, leads the group. Raleigh Music Publishing and Abilene Music are named, along with trusts belonging to the songwriter Luigi Creatore, who co-wrote “Can’t Help Falling in Love,” and members of the Aberbach family, whose publishing business once controlled the rights to songs Presley recorded. Richard Busch, a Nashville lawyer who has spent a decade suing technology companies over music, is lead counsel.
The catalog at issue is the recognizable one. “Can’t Help Falling in Love,” “All Shook Up,” “Don’t Be Cruel,” “Return to Sender” and “A Little Less Conversation” are all in the complaint, alongside deeper cuts such as “Run On” and “Farther Along.” The mix matters to the legal theory. A suit built only on hits invites the argument that the plaintiffs are protecting a handful of valuable properties. A suit built across 274 works argues that the practice was systematic.
The claims run in three directions. The first is direct infringement: the songs were copied without a license and used to train the Claude models. The second is a separate theory about retention, alleging that Anthropic gathered lyrics from across the internet, including web-crawl archives and datasets, and kept them in an internal repository used to develop successive model versions. Under that theory, the library itself is the infringement, distinct from whatever the training run did with it. The third concerns the metadata: the complaint alleges that identifying information was stripped from 247 of the works, which is a claim under a different section of the law than copying.
That structure explains the money. Beyond the $41.1 million in statutory damages, the plaintiffs seek $685,000 for alleged circumvention of technical protections and $6.85 million for the removal of copyright management information, bringing the total to $48.6 million. They also want a permanent injunction and an order that unauthorized copies be destroyed. The complaint requests a jury.
Anthropic disputed the claims. “We disagree with the allegations and intend to defend ourselves,” a representative told the New York Times on Sunday, in the kind of statement companies issue when the litigation is early and the discovery is not.
The backdrop is a docket that has been building for two years. In June 2025, Judge William Alsup ruled in Bartz v. Anthropic that training a model on books the company had lawfully purchased was fair use, and that building a library of pirated copies was not. The case settled last year for $1.5 billion, the largest copyright recovery on record and, in practical terms, a price list for a scanned library. Anthropic has also faced suits from music publishers over lyrics, including a case brought by Concord, which turned on whether a model reproduces protected text when a user asks for it.
Lyrics are harder to defend than books on that question. A novel is long, and a model that memorizes an entire novel is doing something unusual. A chorus is short, memorable and heavily represented in training data, which is why music publishers have been able to point to verbatim output rather than statistical resemblance. The complaint leans on that: it argues that Claude has produced infringing output for songs across the catalog, and that the behavior continued in later model versions after guardrails were added.
Anthropic’s answer will probably separate the two acts. Training on purchased or publicly available material is the activity Alsup blessed. Retaining a library of scraped lyrics is the activity he did not, and the plaintiffs have written their second claim to sit exactly there. The company’s fair-use defense is strongest on the training and weakest on the repository, which is why the complaint spends pages on the repository.
The commercial context is not incidental. Anthropic has publicly filed a registration statement for a listing, disclosing $16.3 billion in revenue for the first half of the year against a substantial net loss, and litigation of this kind belongs in the risk factors of any such document. The complaint makes the point less politely, arguing that the company put a prospective public offering ahead of the rights of the catalog’s owners.
For the music business, the arithmetic is the message. Statutory damages exist so that a small rights holder can enforce a claim against a large user without proving actual harm, and $150,000 per work is Congress’s estimate of what willful conduct deserves. Two hundred and seventy-four works is a small claim for a company of Anthropic’s size, but the template is what catalog owners will copy. Estates and publishers have spent a decade consolidating rights precisely so that a lawsuit can be brought over hundreds of works at once rather than one.
Album catalogs are also becoming easier to police. Rights holders now run crawls of their own, prompting models with lyric fragments and recording what comes back. That evidence is cheap to produce and hard to dismiss, and it converts a debate about the abstract legality of training into a series of specific outputs attached to specific dates.
Anthropic will argue that its models were built to learn patterns rather than to store copies, and that any reproduction is a bug rather than a business model. The plaintiffs will argue that a company which scraped lyrics, stripped identifying information and kept the files has already answered the question. A jury in San Francisco will decide which description fits, and Presley’s catalog will get a second hearing on how it may be used.
