Anthropic hit with a $3B copyright lawsuit over alleged use of pirated music in AI models
 |
| Expect this to keep happening with AI companies. Credit: Jakub Porzycki/NurPhoto via Getty Images |
The world of consumer artificial intelligence has been hit with accusations of theft and piracy for as long as it's existed, and that trend is continuing with Anthropic.
The company behind the Claude chatbot and other AI products was on the receiving end of a $3 billion lawsuit from the music industry late last week, per Reuters. Music publishers including Universal Music Group, ABKCO, and Concord are alleging that Anthropic pirated more than 700 pieces of music (including sheet music and lyrics) they own for use in training Claude. However, the lawsuit also alleges that as many as 20,000 pieces of music could have been infringed upon in this process, hence the massive financial penalty attached to the lawsuit.
Anthropic ought to be used to this by now, having previously been sued by the same music publishers for similar reasons in 2023. Just last year, the company settled with book authors who had accused Anthropic of piracy for $1.5 billion. The same sort of thing has come for other AI companies in the recent past, too, as a German court ruled that OpenAI had violated copyright laws related to music last year.
It remains to be seen exactly how the growing AI industry will reckon with copyright laws as they exist beyond just settling for huge sums of money with every party that sues them. For now, expect lawsuits like this to keep periodically happening until everyone figures out the right way to approach this issue.
Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
Topics Artificial Intelligence
Well, I guess that explains why I stopped being able to explore lyrical analysis. One artist I love has well over 500 tracks, it’s a lot to keep up with lol.
ReplyDeleteClaude is fine with analysis as long as you provide him the lyrics
DeleteYeah idk if you’ve ever tried to load that many lyrical tracks while retaining model fidelity; but it’s beyond my tech abilities.
Delete😅Damn that's a lot to keep up..artist name?
DeleteResponded and text did something weird lol edited my reply though I was wrong but it’s in the 300 range. Tyler Joseph.
DeleteThis comment has been removed by a blog administrator.
DeleteClawd McCode
DeleteIt’s easier to take and ask forgiveness than to ask for permission and not get it. The billions in settlements is just cost of doing business at this scale.
ReplyDeletewhen people build structures they tend to get sued due to unauthorizations if they win they become billionaires, ask Meta, Apple, google, Microsoft, Netflix, coinbase, twitter, how much creative content they obtained through forced Arbitration clauses (user data mining)
DeleteJust what exactly would Anthropic need music stuff for anyways?
ReplyDeleteMajor music publishers, including Universal Music Group and Concord Music Group, filed a $3 billion lawsuit against Anthropic on January 28, 2026, alleging flagrant piracy of over 20,000 works by illegally using BitTorrent and shadow libraries to acquire data.
ReplyDeleteThe suit claims Anthropic executives knew the data sources were sketchy but used them anyway and that the company stripped Copyright Management Information from training data. More information is available on the linked source.
Would be a big fuck up on Anthropic's part if true. They built a capable AI and if this is true, shame on them.
DeleteJust like the authors did https://www.anthropiccopyrightsettlement.com
ReplyDeleteIt is $150K (- legal expenses if any) per artist if they will have and win a trial
ReplyDeleteYeah, but try to talk to claude about music and it's brain goes out the window. Play a game with lyrics: Ask it to give hints as to the 80's song it's thinking of and it will go off the rails. It knows the song Africa, but if you even try to get it to hint at lyrics it was telling me the song was about a guy named Jeffrey in Kenya. It's super funny.
ReplyDeleteMeanwhile, Suno is pumping out music, lyrics, and start personna. But sure... follow the money.
ReplyDeleteYeah and UMG (one of the groups suing Anthropic) also purchased and owns Udio (the other Suno). WTF? They are guilty of the exact charge
DeleteYou've got to be kidding. You can't make this stuff up.
DeleteI can hear the prosecution's opening argument: "Hey Anthropic, you are stealing the valuable artistic labor of generations of musicians. You can't do that-- that's our Fu**ing job! You're crowding the room and... oh yea- you have money"
Artists will get zero dollars out of this too.
ReplyDeleteOf all the companies one might sue about the use of music.
ReplyDeleteOne might successfully argue, by demonstration, that Claude's response to generating lyrics does far more to protect copyrighted lyrics than it does to exploit it.
I tried to get obscure lyrics, well known lyrics, tried to get it to write related spoofs of a song. If you dont train it on copyright material, the free use stuff floating on the internet trains it and, ironically, you might end up with a model that yields much more practical info.
I really hope they consider model outputs as much as inputs.
They want money because an AI heard the song and remembered it? Or am I missing something?
ReplyDeleteFirst, this is the claim made by the record companies.
ReplyDeleteBut Claude wouldn't have the capability to handle music files, so what exactly are they claiming was copied?
Their claim would be about training on the lyrics. I don't know how much of a leg this has to stand on. But recall that anthropic lost (or settled?) the one regarding the pirated books
DeleteOh my, that will sting.
ReplyDeleteLet’s see how the pentagon contract gets resolved now.
These big companies always be like: Pyracy is bad unless we are the ones doing it.
ReplyDeleteYet it can't tell me what Titi me pregunto is about
ReplyDeleteWhy do this??? (Pirating then stealing data?)
ReplyDeleteAccording to the article they have a $183 billion valuation and growing and 3 billion is weather-able…cost of doing business.
DeleteYes but why? i have the max sub and its kinda out of character for anthropic thought they were gonna be better than openai
DeleteWhat do you mean "why"? Not one of these companies got the data they used for training legitimately.. It's all stolen and scraped.. Anthropic is not any different than OpenAI..or Nazigrok or Gemeni or whatever
DeleteYou think that there's a good corporation?
DeleteI believe anthropic trained on 1.8Billion of my tracks how do I sue? Anthropic is always training on music to help their coding abilities it’s terrible :((((((((
ReplyDeleteStuff like this makes me look forward to AI taking over the world.
ReplyDeleteAh yes 13 trillion for anna's archive but 3 billion for AI bros.
ReplyDeleteProbably used less music (20k) to train hence their lower bill compared.
But Meta is allowed to pirate everything on God's green earth with no repercussions.
ReplyDeleteA system that's working as intended, lmao.
It's either the music companies or other techbros sueing tech companies.
ReplyDeleteInteresting. o.o
What's worse is they don't even seed it back. Typical corpos.
Delete"Fair is foul, and foul is fair"
ReplyDeleteIf somebody listen to 10 songs of one genre and proceed to make his own shitty song inspired from those 10 songs can the music companies sue that person?
ReplyDeleteIntriguing accountability in AI is vital. Curious how this plays out?
ReplyDeletemusic industry playbook hasn't changed since napster: sue for billions, settle for licensing deal, call it 'artist protection' while the actual artists see pennies. B is just the opening bid for the toll booth they're building.
ReplyDeleteThe pivot from 'fair use' to 'direct piracy via torrenting' in the legal discovery is the real danger for Anthropic here. (1/2)
ReplyDeleteIf plaintiffs can prove the dataset was seeded from illegal torrents of 20,000+ works, the PBC 'public benefit' shield might not hold up against a $3B statutory damage claim. (2/2)
DeleteTechnicality of 'filtering' is the hurdle. Purging 20k specific works from weights post-training without degrading the manifold is nearly impossible. If the 'transformative' defense fails, Anthropic faces massive technical debt—you can’t 'un-train' neurons on command.
ReplyDeleteThe $3B Anthropic claim is aggressive, but targeting 'shadow libraries' like LibGen is the critical legal pivot. If the court finds founders personally liable for 'active inducement,' it sets a major precedent for how RAG-based LLMs ingest copyrighted sheet music.
ReplyDeleteMeanwhile OpenAI gets away with everything because they have more lawyers and better PR. Pick on the polite ones first, I guess.
ReplyDeleteWe need to be suing OpenAI instead. Anthropic can actually be useful whereas ChatGPT isn’t
ReplyDeleteLegal disputes may affect short-term performance, but the long-term impact on the industry and the company’s ability to adapt is more important.
ReplyDeleteMusic Vs AI just went nuclear.
ReplyDeleteIs this the lawsuit that redraws the rules for creativity?
If artists were paid when their work was used to train AI, it would beat most catalog income. Because training is constant. Models retrain, adjust, improve, repeat. That means steady usage instead of spikes.
ReplyDeleteIt’s the same way Bitcoin mining can provide a steady, predictable load to the power grid. Streaming pays when someone presses play. Training pays because the system depends on the work. That turns art into something durable - a reliable source of income over time. The reality is simple: artists already power these systems. They’re just not being paid for it yet.
This is exactly why on-chain provenance will become mandatory, not optional.
ReplyDeleteTraining data opacity is the risk nobody priced in yet
ReplyDeleteTraining their AI on pirated songs?
ReplyDeleteThat's crazy.
This is exactly the issue. If no one can see what went into the training data, everyone loses, artists and AI companies alike. On-chain provenance feels like the clean, overdue fix before this turns into nonstop lawsuits.
ReplyDeleteThis is exactly why on-chain provenance will become mandatory, not optional.
ReplyDeleteIf you can’t prove your training data, you can’t prove compliance.
Simple as that.
Wtf
ReplyDeleteAI will finally destroy the copyright scam. They will kick and scream but they’re done.
ReplyDeleteThe labels use copyright as a scam. That never was the idea behind copyright and then they will scam all artists as they never will receive their share of this money.
ReplyDeleteFUCK OPENAI TO HELL🖕🖕🖕 #Keep4o
ReplyDelete$3B at stake shows how fast AI hype collides with old copyright empires.
ReplyDeleteThis is a huge moment for AI and copyright law.
ReplyDeleteIf the BitTorrent claims hold up, this isn’t a “fair use” gray area anymore it’s straight-up alleged piracy at massive scale. A $3B+ case like this could set a precedent that forces AI labs to radically rethink how training data is sourced, licensed, and documented going forward.
They will look for a settlement, just like they did a few months ago for 1.5B
DeleteWhy, many other companies have had similar lawsuits about pirating books and nothing happened to them.
DeleteThis isn't true at all. This was already settled under Bartz v Anthropic. Yeah, they'll get hit with piracy, but that's a slap on the wrist compared to copyright infringement, which they won't get hit with.
DeleteBig labels going after AI again.
ReplyDeleteFeels like they are more scared of the tech than the actual piracy.
They are scared
DeleteBoth can be true
Deletethey are lmfaoooo
DeleteThey’ll use the tech with the copyrighted material that they already own
DeleteYes, AI wants to replace them by using them.
DeleteThis comment has been removed by a blog administrator.
ReplyDelete