Elvis Rights Holders Are Right to Force AI Firms to Stop Free-Riding on Genius
The owners of the rights to Elvis Presley’s music have sued the artificial intelligence company Anthropic, claiming the firm violated copyright by using 274 of those songs without any license to train its Claude model. The complaint seeks payments of up to 150,000 dollars per song, a total that could reach roughly 41 million dollars, and describes the material as having been turned into unlicensed derivative works. Anthropic has responded that it disagrees with the claims and intends to defend the case fully.
I read this filing as a necessary line in the sand. If a company can absorb an entire catalog of singular, painstakingly crafted songs and then monetize the resulting model without ever asking or paying, the very idea of ownership in music collapses into a polite fiction. The labor, the risk, the decades of cultural weight attached to those compositions do not vanish simply because a machine can now ingest them at scale. In my judgment, that ingestion is not neutral study; it is appropriation dressed up as progress.
We face a clear tradeoff. On one side sits the convenience and speed of training systems on whatever data can be scraped. On the other sits the principle that creators and their successors control how their work is exploited. I come down firmly on the second side. Conditional reasoning makes the stakes plain: if every future model is allowed to treat iconic catalogs as free raw material, then the economic signal that once rewarded distinctive songwriting weakens until it barely registers. Why invest years refining a melody or lyric when an unlicensed system can remix its essence overnight and sell access to the result?
Protection here is not nostalgia. It is the recognition that some works carry a density of invention that later tools still lean on. Allowing that lean without compensation turns the original achievement into an unpaid subsidy. I find that outcome ethically thin. The request for substantial statutory damages is, to my mind, proportionate precisely because the alleged use was systematic rather than incidental.
(And yes, I admit a soft spot for the idea that certain songs feel almost sacred in their construction—hardly a legal argument, yet it sharpens my sense that careless reuse cheapens them.)
Later stages of the dispute will turn on doctrine, but the deeper question is already visible. Do we treat the training corpus as a commons that anyone may drain, or as a set of bounded properties that require permission and price? I choose the latter. When the boundary is ignored at industrial scale, the only coherent response is to restore it through the courts. Anything less invites a race to the bottom in which the most aggressive scrapers win and the originators are left holding empty rights.
If the model truly could not have reached its present capability without those 274 songs, then the value extracted from them belongs, at least in part, to the rights holders. Refusing to acknowledge that transfer is not innovation; it is extraction. I see no persuasive moral offset that justifies zero compensation. The lawsuit, therefore, is not an attack on technology. It is a defense of the conditions that make distinctive creation worth undertaking in the first place.
Should courts ultimately agree, the precedent will simply restate an old bargain in a new setting: take the work, pay for the work. That bargain has never been perfect, yet abandoning it now would be worse. The alternative—an AI economy built on unconsented absorption of the most enduring catalogs—strikes me as both unfair and shortsighted. Better to insist on licenses, however imperfectly priced, than to pretend the music somehow donated itself to the training run. The rights holders are acting exactly as stewards should: protecting what they hold so that its value is not quietly dissolved into someone else’s product.
