Anthropic Settles Major AI Copyright Suit Brought by Authors (bloomberglaw.com) 24
Anthropic reached a settlement with authors in a high-stakes copyright class action that threatened the AI company with potentially billions of dollars in damages. From a report: In a Tuesday filing in the US Court of Appeals for the Ninth Circuit, both sides asked the court to pause all proceedings while they finalize the deal. The parties signed a binding term sheet on Aug. 25 outlining the core terms of a proposed class settlement to resolve litigation brought by authors.
"This historic settlement will benefit all class members," said the authors' counsel, Justin Nelson of Susman Godfrey LLP. "We look forward to announcing details of the settlement in the coming weeks." The case is one of several copyright actions brought against AI developers in courts around the country. Judge William Alsup of the US District Court for the Northern District of California had allowed the class action to proceed for authors whose books were contained in two pirate databases Anthropic downloaded.
"This historic settlement will benefit all class members," said the authors' counsel, Justin Nelson of Susman Godfrey LLP. "We look forward to announcing details of the settlement in the coming weeks." The case is one of several copyright actions brought against AI developers in courts around the country. Judge William Alsup of the US District Court for the Northern District of California had allowed the class action to proceed for authors whose books were contained in two pirate databases Anthropic downloaded.
Wrongdoing (Score:5, Interesting)
I'm guessing that part of the settlement lets Anthropic off the hook without having to admit wrongdoing. So Anthropic won. It gets to keep the spoils of its crimes, while paying a mere tiny fraction of what it was on the hook for.
Re: (Score:3)
This outcome was predicted by myself and others right here on Slashdot. And probably everywhere else, too.
There is simply too much interest, and far too much money, behind AI for something as petty as laws or justice to stand in its way.
Remember ye well the golden rule: he who hath the gold, rules.
Re: Wrongdoing (Score:2)
Wasn't this the suit where the finding was anyway that the only copyright violation was pirating the books to be used for training from pirate sites, and that it would all have been fair use if they had actually legally obtained copies of the books? (I'm probably in minority about this, but this is also the way I argue it should beâ"as judge Alsup said, it's hard to think of a more transformative use than training LLMs, and there has never been a generic right to control or profit from every way your c
Re: Wrongdoing (Score:2)
Also, this means that there actually would be quite good incentive for the authors to settle. What Alsup found was really not what the authors hoped the law would be. Settling minimizes its impact and precedential value.
CRIMES? (Score:1)
I'm guessing...
Yes, based on nothing.
So Anthropic won. It gets to keep the spoils of its crimes...
You're confusing criminal actions with civil liability. Back to law school you go.
...while paying a mere tiny fraction of what it was on the hook for.
Stil making up stuff? You have no idea what any of it was worth, what Anthropic was on the hook for, or what "tiny fraction" ALL PARTIES AGREED ON.
But hey, it's your ass, let it keep talking.
Re: (Score:3)
Yes, based on nothing.
Based on not knowing the terms of the settlement. The story didn't say. Given how often settlement terms are kept confidential, we can draw a reasonable conclusion.
You're confusing criminal actions with civil liability. Back to law school you go.
U.S. copyright infringement can be civil or criminal. Anthropic's behavior falls into the requirements for criminal copyright infringement. See "17 U.S. Code 506 - Criminal Offenses". That is exactly Anthropic's business model, and what Judge Alsup said would likely be their undoing if the case went to a Jury trial.
You have no idea what any of it was worth....
Quite the opposite. Anthropic's
Re: (Score:2)
Let's assume Anthropic is a business. It's even likely they would be happy to pay a fair licensing fee based on profits. It certainly would be easier. But, they will probably be hit with lawsuit after lawsuit so to survive, they need to pay as little as possible per suit.
Now, there are the lawyers. They might find a group of tens of thousands of people to represent. Since they established the group, there is no leader or clear representative for the group, so the lawyers cho
Re: (Score:2)
Lawyers have always been evil. That's never going to change. Anthropic and the AI industry is a new evil. That bothers me more.
Re:Wrongdoing? (Score:2)
Well, if they can get away with paying a few cents licensing to use "The Stand" to train a model, and then that model can regurgitate verbatim any passage from the book, isn't that kinda what all that jargon about "... not reproducing, storing, or entering into a retrieval system, or transmitted in any form, or by any means"
So... if they can get away with having their "A.I" regurgitate passages from a book on a paid service, couldn't one argue that Anthropic just negated copyright?
Shouldn't it totally fine
Meanwhile in the ToS (Score:2)
All of these companies tell you that you can't even use the output to train a competing model.
Re: (Score:3)
And no user cares about that.
It's all disclaimers. Your model went shit? We told you not to train on our output!
They also know that models can be jailbroken and abliterated. But as long as the model by default refuses to write smut, they can blame others when they make the model write smut nevertheless.
In the end, they don't want the people to point their fingers on them when something goes wrong.
Authors fucked by the lawyers? (Score:2)
I can't see the authors getting anything meaningful from this if the settlement is this quick.
I think a lot of authors will want some meaningful money, not a pittance just so the lawyers can make off with a fat paycheck for little work.
As one of the victims here... (Score:2)
I saw a chatgpt conversation directly copying equations from my book - The equations for computing SCC from pattern counts which appear only in my book as far as I am aware.
I haven't seen them or any lawyers or publishers offering me money for this blatant copying.
Re: (Score:2)
Citing equations is no copyright issue. Every other author could cite that equation. For scientific conduct they should also cite you as original author, but that is unrelated to the copyright question (and rather a plagiarism question which is a different deal). In particle even long equations are not protected, as facts can't be copyrighted. An equation is literally a statement "X equals Y is true" (sometimes X is lesser than Y and so on. Always being "here is a fact derived to be true").
Re: (Score:2)
But a body of text, taken from the book, including the equations in exactly the format in the book (when there are 32 ways you could choose to arrange the terms if you were doing it from scratch). That's not ok.
Citing "So and so in the book xyz derived the following equation ...." would be fine. Copying and pasting with neither attribution nor rephrasing to make it fit the narrative of the chat is copyright infringement. More importantly they clearly have absorbed my book into their model without paying for
Re: (Score:2)
I agree. If something is indistinguishable from copy&paste it does not matter how layers of indirection were used for it.
I only never saw an reproducible example of such a claim. Most the time people put some of their text in and then got some plausble extension that matches to a certain extend their own text. I mean if I ask someone to complete the sentence "Harry Potter is a" I do not need a language model to predict the words with the highest probabilities. And if a model completes the lord's prayer
How do I get my share? (Score:2)
I (genuinely) have a couple of hundred books of mine that are/were "pirated into" the LibGen database... so... does anyone know how I can go about claiming my settlement cash?
You missed the point (Score:2)
Skimming the comments, y'all mised the point: Anthropic made $5B in revenue, while loosing *many* billions in operating costs.
So, how much longer will they be in business, before they collapse (and vulture capital pulls out)?
Also, given the reported in the media emails that *said* that the CEO *chose* not to pay originally "because it was too complicated", they were looking at a 100% chance that they'd be found liable.
Collapse of the AI bubble, coming soon.