Music Publishers Reach Deal With AI Giant Anthropic Over Copyrighted Song Lyrics (hollywoodreporter.com) 8
An anonymous reader quotes a report from the Hollywood Reporter: A trio of major music publishers suing Anthropic over the use of lyrics to train its AI system have reached a deal with the Amazon-backed company to resolve some parts of a pending preliminary injunction. U.S. District Judge Eumi Lee on Thursday signed off on an agreement between the two sides mandating Anthropic to maintain existing guardrails that prevent its Claude AI chatbot from providing lyrics to songs owned by the publishers or create new song lyrics based on the copyrighted material.
In a statement, Anthropic said Claude "isn't designed to be used for copyright infringement, and we have numerous processes in place designed to prevent such infringement." It added, "Our decision to enter into this stipulation is consistent with those priorities. We continue to look forward to showing that, consistent with existing copyright law, using potentially copyrighted material in the training of generative AI models is a quintessential fair use." [...] Under the agreement, Anthropic will apply already-implemented guardrails in the training of new AI systems. The deal also provides an avenue for music publishers to intervene if the guardrails aren't working as intended.
"Publishers may notify Anthropic in writing that its Guardrails are not effectively preventing output that reproduces, distributes, or displays, in whole or in part, the lyrics to compositions owned or controlled by Publishers, or creates derivative works based on those compositions," the filing states. "Anthropic will respond to Publishers expeditiously and undertake an investigation into those allegations, with which Publishers will cooperate in good faith." Anthropic has maintained in court filings that existing guardrails make it unlikely that any future user could prompt Claude to produce any material portion of the works-in-suit. They consist of a "range of technical and other measures -- at all levels in the development lifecycle -- that aim to prevent users from simply prompting Claude to regurgitate training data," said a company spokesperson. The court is expected to issue a ruling in the coming months on whether to issue preliminary injunction that would bar Anthropic from training future models on lyrics owned by the publishers.
In a statement, Anthropic said Claude "isn't designed to be used for copyright infringement, and we have numerous processes in place designed to prevent such infringement." It added, "Our decision to enter into this stipulation is consistent with those priorities. We continue to look forward to showing that, consistent with existing copyright law, using potentially copyrighted material in the training of generative AI models is a quintessential fair use." [...] Under the agreement, Anthropic will apply already-implemented guardrails in the training of new AI systems. The deal also provides an avenue for music publishers to intervene if the guardrails aren't working as intended.
"Publishers may notify Anthropic in writing that its Guardrails are not effectively preventing output that reproduces, distributes, or displays, in whole or in part, the lyrics to compositions owned or controlled by Publishers, or creates derivative works based on those compositions," the filing states. "Anthropic will respond to Publishers expeditiously and undertake an investigation into those allegations, with which Publishers will cooperate in good faith." Anthropic has maintained in court filings that existing guardrails make it unlikely that any future user could prompt Claude to produce any material portion of the works-in-suit. They consist of a "range of technical and other measures -- at all levels in the development lifecycle -- that aim to prevent users from simply prompting Claude to regurgitate training data," said a company spokesperson. The court is expected to issue a ruling in the coming months on whether to issue preliminary injunction that would bar Anthropic from training future models on lyrics owned by the publishers.
And the Copyright Enshittification begins (Score:3)
Re: (Score:1)
Currently huggingface is full of uncensored community-tuned LLMs. Go and make a decentralized backup before huggingface starts cleaning as well.
I don't think they currently have plans, but when some XXXmedia site writes the next clickbait article about how uncensored LLM can create songs based on existing lyrics, the enshittification may come soon.
Re: (Score:2)
You realize that would destroy Free/Open Source Software, right?
Because what protects the Linux kernel among other things, is copyright. Yes, copyright is what makes the GPL actually enforceable. You can completely ignore the GPL, and use Linux, but then Linux will be licensed to you under copyright "all rights reserved". If you want to m
ANTICOPYRIGHT industry ruins online cultrue. (Score:1)
I'll purposely not go into why copyright exists, provides incentives to creators, etc. Intead:
The industry created to "police" (they are law enforcement), "fight" (they are not any military force), "remove" (they are trying to be a de-facto censorship organization) and "FILE SUITE" (their one true calling) is the bane to the existence of any work that makes use of any other work.
The problem we have in our society of today is allowing these copyright-maximalist to pretend to be part of the society in which
Re: (Score:2)
As long as they fuck with Scam Saltman at least they are doing something useful.
Closed models should license their training set. Fair use should only be possible for open models.
Guess how much (Score:3)
the artists will get.
Nothing interesting decided yet (Score:2)
Digitalmusicnews wrote an article with an actual PDF of the judgement, can't even copy paste from scribd any more without an account.
https://www.digitalmusicnews.c... [digitalmusicnews.com]
The actual interesting rulings are still up in the air : "In particular, the Parties continue to dispute, and this stipulation does not resolve or affect, Publishers’ request in their Motion for Preliminary Injunction that Anthropic refrain from using unauthorized copies of Publishers’ lyrics to train future AI models."
So computer translated lyrics are OK for training? (Score:2)
Conjecture: It should be possible to split popular song lyrics and how they are sung into multiple parts, then create nonsensical roughly equivalent patterned song lyrics using an AI. Then take that nonsensical lyrics and use another poetry trained AI to make them nearly coherent.
1. Meter
2. Rhythm
3. Repetition
4. Word emphasis per line and location in the lyrics
1) Line by line, word by word swap song lyrics for roughly equivalent sounding words for the meter and rhythm
2) Train an AI on that data set
3) Use t