Elvis Presley's rightsholders sue Anthropic for $41M over 274 songs

Two bills came due this weekend: one for the songs that went into a model, one for the weather models that got outperformed by one. The Elvis estate wants $41 million. The forecasters are rethinking their defaults.

Authentic Brands Group, which controls Elvis Presley's music and likeness, sued Anthropic on Friday for training Claude on 274 of Presley's songs, and it wants $150,000 for each one. The complaint, filed as ABE EPE LLC v. Anthropic and first reported by Billboard, totals roughly $41 million — and it is one of the first major copyright cases brought by an artist's estate rather than by a label, a publisher or the RIAA.

The catalogue in question is not obscure. The suit names "All Shook Up," "Can't Help Falling in Love" and "Don't Be Cruel" alongside lesser-known tracks, and it argues that Anthropic's use produced "unlicensed derivative works" rather than anything transformative. Lead attorney Richard Busch — who worked the Marvin Gaye estate's case against Robin Thicke and Pharrell over "Blurred Lines" — wrote that the plaintiffs "are deeply protective of these iconic songs" and that the defendant "has built its entire business model based on copyright infringement at a scale never seen before," according to Rolling Stone.

The most interesting part of the filing is not the training claim, which is now a familiar genre. It is the argument that Anthropic kept reproducing the material after it was sued. The complaint alleges that when Claude was recently asked to recreate specific songs "for a modern audience," the model complied and then described its own output as "edging past 'inspired by' into reproducing the copyrighted song." That quotation is the plaintiffs' strongest asset, because it attacks the guardrails directly: if the filter was installed after the first wave of music litigation and still leaks, the case for willful infringement — the multiplier that gets you to $150,000 per work instead of the $750 statutory floor — gets much easier to argue.

Anthropic's position is that the claims are wrong on the law and it will defend itself, it told the New York Times. That is the standard response, and it is also the substantive one. Training on purchased or scraped recordings and emitting near-identical outputs are legally different acts, and the company's best defense runs through the second: a model that memorizes is a bug to be fixed, not a business model. The plaintiffs' answer is that they have receipts showing the memorization is not incidental.

The money is almost beside the point. Forty-one million dollars is a nuisance against a company preparing a public listing, and everyone in the room knows it. What the Elvis estate is buying with the filing is a precedent that a rightsholder can reach a model lab directly, without waiting for a label trade group to negotiate a licence on their behalf — and a signal to every other estate sitting on a catalogue with a lawyer on retainer. The label suits have been running since the summer, when Sony Music and Warner Chappell accused Anthropic of a "brazen campaign" of lyric piracy — we covered that filing in Sony and Warner sue Anthropic for 'brazen' lyric piracy — but those cases are about the industry's terms. This one is about whether an estate can set its own.

The counter-argument worth taking seriously: statutory damages at the top of the range are reserved for conduct a defendant knew was wrongful, and Anthropic can point to a licensing posture that has changed over two years, filters added under protest, and an output complaint that is a screenshot of a chatbot improvising. A judge may read that as evidence of good-faith iteration rather than a confession. The estate's case is stronger on rhetoric than on the record so far, and rhetoric does not survive summary judgment.

Google DeepMind's hurricane model called Hurricane Isaias' landfall while the conventional physics models had it going somewhere else — and it was right. In early October, the standard models had the storm drifting further west toward southeastern Louisiana. DeepMind put landfall on the western Florida Panhandle from the first run. Isaias came ashore near Destin as a Category 2 on October 9, killing at least four people and cutting power to more than 390,000, as WUSF's Jessica Meszaros reported in a piece carried by NPR.

The track record is the story. The National Hurricane Center only started folding the DeepMind model into its official forecasts last year, and by the end of the 2025 season it was already the best performer on both track and intensity — an unusually fast climb for a model type that normally takes years to tune. "That model has been, by far and away, the best performing computer model that we've ever seen. Period," Michael Lowry, a Miami meteorologist who has advised FEMA and the hurricane center, told NPR. Brian McNoldy of the University of Miami put the surprise more plainly: "This pretty much came out of the gate in first place, and that's a pretty impressive feat."

The caveats matter, and the hurricane center's director supplied them. DeepMind won the track. It did not beat the center's blended forecast on intensity — how strong the storm gets — and NHC director Michael Brennan describes the operational answer as a consensus blend of AI and physics models, with humans providing the continuity that keeps a six-hourly forecast from lurching between runs. "Every storm has different forecast challenges," he said. Nor is the model transparent: it learned from thousands of past storms rather than from the governing equations of the atmosphere, so nobody can fully explain a hit or a miss. "We like to know why you're getting the right answer," McNoldy said. "And I feel like we don't."

For readers outside meteorology, the useful number is not the hit rate — it is the update cadence. AI forecast models improve on a cycle of months or weeks, against an annual refresh for physics models, which is why Lowry describes a year of AI progress as worth a decade of the old pace. Intensity errors are already about half what they were in the Katrina era, per Brennan. We looked at where that trajectory was heading when the newest iteration shipped — Google's WeatherNext 3 trades the six-hour lag for hourly forecasts — and the Isaias result is the first clean operational test of it.

What to watch: whether the hurricane center's end-of-season verification shows the AI edge holding on intensity, not just track — that is the difference between a better steering model and a replacement. On the music side, watch for the first estate that cites this complaint rather than files its own.

If a forecast is right for reasons the forecaster cannot explain, should it carry the same weight as one you can audit? Write to us via our contact page.