Trained to Imagine
Rendering
Creative Intelligence
AI Production / Hollywood / Legal

Anthropic sued over Elvis Presley recordings used to train Claude

Source: The Hollywood Reporter ·

Authentic Brands Group and other Presley rights holders sued Anthropic in federal court in San Francisco, alleging 274 songs were used without permission to train Claude and that the model reproduced them on request. They seek statutory damages of up to $150,000 per song and an injunction; Anthropic said it disagrees and will defend the case.

TTI Takeaway

The training-data fight is no longer confined to image and video models. Estates with a single, famous catalogue are a different kind of plaintiff from a label group — narrow claims, easy to explain to a jury. For anyone producing music-led work, it is another reason to ask what every model in the chain was trained on, including the ones that only write.

Analysis

This case is notable less for anything it has decided — it has decided nothing yet — than for the shape of the claim and the kind of defendant. Most music-industry suits against AI companies have targeted audio generators, and most have come from publishers or labels holding very large catalogues. Here the plaintiffs are Authentic Brands Group and other entities holding Presley rights, the defendant is the maker of a language model, and the complaint is built around a defined list of 274 songs, including “Can’t Help Falling in Love,” “A Little Less Conversation” and “Don’t Be Cruel.”

As The Hollywood Reporter describes the filing, the allegations run along two lines. The first is that the songs were used to train Claude without permission. The second concerns output: the complaint alleges that Claude reproduces protected material, and quotes an exchange in which the model, asked to recreate specific songs “for a modern audience,” is said to have described its own results as “landing very close to the originals.” The plaintiffs argue this happened even after guardrails were introduced following earlier lawsuits. Lead attorney Richard Busch wrote that the defendant “has built its entire business model based on copyright infringement at a scale never seen before.” Those are the plaintiffs’ characterisations. Anthropic told THR that it disagrees with the claims and intends to defend the case fully. None of it has been tested in court.

The filing does not arrive in isolation. THR notes that Anthropic was previously sued by Universal Music Publishing, Concord, BMG and others over lyrics, and that Round Hill sued both Suno and Anthropic in August. On the audio side, the feed has followed the GEMA ruling against Suno in Germany in July and the Universal and Sony amendment adding stream-ripping claims against Suno in August. What sets this one apart, from a production point of view, is that the tool at issue writes text. The dispute is about lyrics and song structure, not sound files.

That is where it touches day-to-day work. Music-led productions — music videos, commercials built around a licensed track, lyric-driven campaigns — now routinely use language models for treatments, scripts, lyric rewrites and social copy. The quiet assumption that text tools sit outside clearance is what this suit puts in front of a court, whatever the eventual outcome. In our view the practical response is procedural, not a ban. If a model is asked to rework a known song’s lyrics for a spot, the result should go through the same publishing and sync review as if a copywriter had done it from the record. And productions should keep a simple log of which models touched which creative elements, so the answer exists when a client’s legal team asks.

The open questions are the ones to watch. The plaintiffs seek statutory damages of up to $150,000 per song and an injunction. How a US court treats the output allegations, as distinct from the training allegations, may matter more for working creatives than the headline figure, because output is the part a production can actually control. It is also worth watching whether any of the earlier publisher cases resolve in ways that set terms for the industry before this one gets far. Until then, the sensible position is to treat every model in the chain, including the ones that only write, as something whose output needs the same scrutiny as any other contributor’s.