Anthropic and 20,000 stolen songs: music publishers want $3 billion

Imagine walking into a store for bread and being handed a bill for the entire supermarket's inventory, including warehouse stock. That's roughly the situation Anthropic finds itself in, as music publishers have decided to make the company a cautionary tale. The initial lawsuit, which concerned a modest five hundred compositions, suddenly ballooned to catastrophic proportions. Now the creators of Claude are accused of stealing 20,000 works, and the sum of claims has reached a staggering three billion dollars. The music industry clearly refuses to tolerate its intellectual property being used as free fuel for neural networks.

Music giants like Universal Music Group, Sony Music, and Warner Chappell have never been lenient with those attempting to profit from their content without permission. We remember the Napster era and endless lawsuits against torrent trackers, but now the adversary looks far more serious and wealthier. Publishers argue that Anthropic engaged in "egregious piracy," feeding copyrighted song lyrics to its neural network. This is not about random coincidences, but systematic use of massive datasets to train a commercial product that now competes with the authors themselves.

Why did this strike come now? Anthropic had spent considerable time carefully building an image as an 'ethical' and safe alternative to OpenAI. While Sam Altman clashed with newspapers and artists, Dario Amodei and his team emphasized their commitment to principles and responsibility. However, for music label lawyers, there's no difference between 'safe' AI and any other if it reproduces a Taylor Swift song lyric on the user's first request. The sharp increase in the list of claims from 500 to 20,000 songs suggests that during litigation, new details emerged about the scope of the model's training. It appears Anthropic hoped no one would notice an extra ten thousand songs in the training dataset.

Three billion dollars is not just a fine; it's an attempt to question the very possibility of LLMs existing in their current form. Music bosses want to set a precedent that will force tech giants to pay for every byte of copyrighted content. The legal concept of 'fair use,' which neural network developers so rely on, is cracking at the seams. Publishers insist that creating a commercial model capable of generating text in the style of famous authors has nothing to do with fair learning or academic citation. It's business built on others' work, and labels want their cut.

For Anthropic, this lawsuit could be an existential threat. Investors who poured billions into Claude's development hardly expected that a significant portion of those funds would go toward settling copyright infringement claims. The situation is complicated by the fact that proving 'innocence' in a world where model weights are a black box is virtually impossible. If Claude can quote a verse from an '80s hit, that hit was in its training dataset. And for that, full payment will be demanded. The entire AI market is now holding its breath, as the outcome of this case will determine the rules of the game for years to come.

The bottom line: The music industry has gone on the offensive, and Anthropic will have to prove that their 'ethical AI' isn't built on a foundation of pirated lyrics. Can the company survive if the court recognizes each of the 20,000 songs as a separate violation?