Authentic Brands Group, which owns the rights to Elvis Presley’s music, has sued Anthropic in a case filed on Friday. The claim is that the company trained its Claude model on 274 Elvis songs without paying for a licence.
A spokesperson for Anthropic told the BBC: “We disagree with the claims and intend to defend this case fully.”
The songs named in the complaint include All Shook Up, Can’t Help Falling in Love, A Little Less Conversation, Don’t Be Cruel and Return To Sender. Authentic Brands says these were used to create “unlicensed derivative works”, and asks for payments of up to $150,000 for each infringed song — about $41m in total.
The language is doing work
The filing is written for effect. The catalogue, it says, “contains some of the most groundbreaking, revolutionary, influential, and enduring musical compositions in human history”, and the plaintiffs are “deeply protective of these iconic songs”. Anthropic, it alleges, “has built its entire business model based on copyright infringement at a scale never seen before”.
Courts are not generally moved by adjectives. But one passage in the complaint is different in kind, because it is not the plaintiffs’ characterisation at all.
The quote from the model
Authentic Brands argues that Anthropic’s mitigations did not work: “Even after the Defendant supposedly put up ‘guardrails’ after being sued, in one of its latest iterations, Claude continues to infringe.”
It then quotes the model. Asked to recreate specific songs “for a modern audience”, the complaint says, Claude did so and then said of its own output: “These renders keep landing very close to the originals: same structure, same hooks reused with light rewording, and full sets of verses/choruses/bridges.” The model added that this was “edging past ‘inspired by’ into reproducing the copyrighted song, even when the words are swapped out”.
Whether a chatbot’s self-assessment is evidence of anything is a question for a judge. A model has no knowledge of its own training data and no standing to concede a legal point; it is producing text that reads like an admission. But as a demonstration that the output resembles the input closely enough for the system itself to describe the resemblance, it is a neat piece of lawyering — and it sidesteps the usual defence that a model learns patterns rather than copying works.
The third suit in ten months
This is not an isolated action. Anthropic was accused last week of using lyrics owned by the music company Reservoir. In January it was sued by Universal Music Group. Both cases remain ongoing.
And the company has already paid once at scale. Last year it agreed to pay $1.5bn to settle a class action brought by authors who said their work had been taken to train its models — described by the authors’ lawyers as the largest publicly reported copyright recovery in history.
That settlement is the reason the Elvis figure looks modest. At $41m, this claim is less than 3% of what Anthropic has already shown itself willing to pay to make a training-data case go away.
Why the timing matters
The suit landed in the same week Anthropic released the 335-page filing ahead of its stock market debut. Pending litigation has to be disclosed to investors, and a listing expected to value the company above $2 trillion gives every rights holder with a catalogue a reason to file now rather than later: the defendant is about to be both very visible and, on paper, very rich.
For readers here, the useful point is not the amount. It is that “we trained on publicly available material” is being tested in court, repeatedly, by plaintiffs with the resources to see it through — while the models in question are already embedded in products people pay for. Anyone building on top of a commercial model is building on an input whose licensing status is still being litigated.
Related: a Canadian province suing OpenAI, where the legal question is harm rather than copyright.
