If this could be upheld (though unfortunately it won’t) then it could completely end the generation of new models and the companies that make them. The models that are already out would never go away and might as well be available for people to run locally but if they want new ones it should be done for the collective good with oversight. I.e. state run.
But again, just wishful thinking on my part.
In fact, as Copyright Lately‘s Aaron Moss points out, even the AI industry called the latest ruling a win, with tech industry group Chamber of Progress senior director of AI Adam Eisgrau tweeting that the ruling “implicitly confirms that highly transformative gen AI training to produce a hugely multi-purpose model with substantial public benefit is likely fair use!”
“We stole even more, so that makes it legal!”
So it is still legal for me to pirate anything I want as long as i train a small ass AI model on the side with it?
Remember to clone a repo for an AI for plausible deniability.
When ask why your consuming the copyrighted material simply say you’re in the quality control phase.
This would be a separate issue from piracy
This is more about what you can DO with something you legally purchased. Think film rights. If you want to use a song in your movie, you can’t just go to Tower Records and buy the CD. Buying the CD gives you a right to listen to that CD, but it does not give you the right to use it in your movie.
Likewise, this court argues, having legal access to a work does not give you the right to feed it into an LLM as training data.
It’s a bit like the Muppet Show. The Muppet Show was a bunch of puppets singing along with musical pop stars of the day. But because the show was produced before VHS and DVD were a thing, none of the artists had given permission for them to re-release the show on DVD. So re-releasing the Muppet Show was a tedious process of negotiation with every rights holder.
But I absolutely hear you and am currently training a small LLM on my favorite TV, movies, and music. Have I started the actual training yet? No. I don’t see the need until I’ve filled up this 15TB HDD with training data so I can do it all in one pass.
Likewise, this court argues, having legal access to a work does not give you the right to feed it into an LLM as training data.
Seems more narrow than that. If your “AI” just copies the input almost verbatim, then it’s not “transformative”.
Unfortunately they still seem to regard the big players’ consumption as “transformative” enough to not count.
>“How influential the latest appeals court ruling will be is debatable. The court found that Ross had effectively copy-pasted Thomson Reuters’ “headnotes,” or brief editorial summaries of legal issues, verbatim for its legal search engine, a practice that isn’t difficult to separate from fair use.”
>“In fact, as Copyright Lately‘s Aaron Moss points out, even the AI industry called the latest ruling a win, with tech industry group Chamber of Progress senior director of AI Adam Eisgrau tweeting that the ruling “implicitly confirms that highly transformative gen AI training to produce a hugely multi-purpose model with substantial public benefit is likely fair use!””
Looks like the company that got sued just copy pastaed protected content without making any type of change to make it fair use.
Not the “win” anti-AI people were looking for and the article seems to be misunderstood by most.
Seems as long as AI models use some form “transformative” method to train their AI, then it’s fair use.
Let them fight
I want both sides to lose and for it to really hurt.
It would be best if they destroyed each other, but perhaps that’s overly optimistic
I ficking hope so. Either all profits aren’t privatized and distributed socially or blow the whole thing up.
The headline isn’t correct. This ruling is about a specific element of Fair Use, not a blanket declaration that copyrighted data cannot be used to train models as the headline suggests.
Fair Use requires some conditions be met. The one that wasn’t met here was that the work you’re creating can’t compete with the copyright owner.
From the Reuters article:
“Ross took the headnotes to make it easier to develop a competing legal research tool,” the judge said. “So Ross’s use is not transformative.”
In other cases, for example the Author’s Guild lawsuit against Meta/Anthropic/etc courts have ruled that using books to train LLMs is fair use because LLMs and books don’t compete.
They were required to pay for the books, but not prevented by copyright from using them to train a model.
There’s nothing novel happening in this case from a legal perspective. Maybe this is the first case where the defendant didn’t clear the elements of fair use but those elements predate AI by decades and have been uses in thousands of cases.
I hope Studio Ghibli sue them for a bazillion dollars
Still doesn’t prevent data mining of forums for, “training.” Still doesn’t give people who’s contributed on forums a check for being of value to big tech. If they cared at all they’d be paying us money to train AI until it’s established. Google is offering points for data rather than money (https://deviceusagestudy.google/signup/invitecode) I only know that because I got a flier in the mail which means they’re desperate to get people to sign up.
As much as I hate AI, current copyright laws are out of control and are extended well beyond what they should be used for.
I agree, but I fucking hate that it’s AI that’s breaking the concept. First of all, it’s not like it’s a righteous fight, whoever wins, we all lose. Second, as much as I think copywrite is abused like hell, I still think the idea, one closer to the original, has merit and does provide crucial protections for individuals.
Sadly the fucking plagiarism machine and the anti-human dirtbags running it have found the absolute worst possible way to break the system. Ultimately resulting in nothing for us plebes that copywrite should protect and all the benefits that the shit copywrite coroporate power grab produced for them.
AI is way more out of control than copyright.
There’s simply no way to justify what the AI compies have been doing. As that one Microsoft exec said, it’s the “largest theft of labor” in modern history
Wouldn’t be funny if disneys fuckery did in AI?
I mean Disney’s lawyer game is top notch. I would love to see the mouse vs AI battle.
As long as we live under capitalism, copyright protects small artists, writers, and creators as much as it protects large shitty corporations who weaponize it.
I’d love to be rid of all intellectual property, but we have a lot to fix before that can become reality without punishing people who rely on it.
I’m down with some huge reforms though.
Its actually mainly used to bully small creators so they can have precedent to go after larger ones.
No, not at all. Do you have any idea of how expensive it is to bring a lawsuit? What you are saying about protecting small artists is pure propaganda. Furthermore, this only matters for commercial artist as copyright is never used in 99.999% of real life cases
We will never move forward as long as we allow all of our modern culture and technology to be locked up by corporations through imaginary property rights.
It also costs money to register a copyright. And if it’s not registered when the copyright infringement happens, statutory damages don’t apply and you need to prove actual damages instead.
I hate them too but AI is more of a threat to all of us. Any chance at stopping the madness is good enough for me.
Betteridge’s law of headlines in full effect.
For those (like me) who don’t know what Betteridge’s Law of Headlines is: “Any headline that ends in a question mark can be answered by the word no.”
Now I feel bad for basing my programming career on material I apparently “stole” from books and other copyrighted sources. Hell, sometimes I even straight-up copypasted example code. I blame society.
I don’t care if copyright holders win I just want AI to lose.
I’d kinda prefer both lose…
Yeah that is the ideal outcome.
When peasants were able to copy ‘intellectual property’ for personal benefit and enjoyment, it was a massive crime. When the oligarchs can steal - everything - for the slight chance of paying less in wages, it is an innovation.
I think it is pretty much certain the model companies escape any penalty for their mass theft of IP. There will be some settlements and token penalties that sound big on paper, but are nothing compared to the market valuations.
It’s almost like rules are just made up
Not really. We had collage and training any kind of AI was fine before all this. There’s blood in the water and copyright companies can smell it.
OpenAI is happy to give them a piece of the pie if that means they are the only ones allowed to sell it to consumers.
You are right about the penalty being pennys compared to their profits. It’s not meant to punish them but help them build a moat, the one that got destroyed by China a few months ago.
I’m cynically believing that this is the beginning of enforcing a copywrite use tax; when anyone generates something that is visual or audible.
The tax will be collected by the government and put into a slush fund that mostly goes to corporations and wealthy, while ostensibly supposed to go fairly to all creators
I’m guessing it’s going to be a system where a company like Spotify doesnt yet you upload generated music unless it has the metadata from the “legal” music generators so they can automatically send most of the profits to the shareholders.
Universal already owns Udio, Suno will follow soon. I wish it was an actual tax.
This is your reminder that in 2026 absolutely ZERO court rulings matter long term except the Supreme Court, which is captured for life by conservatives.
No matter who wins the midterms or even the next election, that is with us for the rest of most of our lives.
What a time to be alive.
This is why the court MUST be packed. FDR threatened to do it 90 years ago, and the Dems better do it this time or else face another Trump again.
Only in the US. EU has teeth and occasionally uses them.
It very much is not. It is in effect until the end of the United States. This will be long before most of us die.
The future will be getting very bad very fast and most nation states will collapse in a few decades.
So there’s that I guess.
I think you are DRASTICALLY underestimating how durable fascist regimes are. Without external intervention they can continue for DECADES. Societies don’t just collapse when people get fed up.
I don’t mean when people are fed up. I mean when millions are starving to death and the firestorms are ravaging our croplands. When the hurricanes give us a Katrina level disaster 6 times a year. When it becomes literally impossible for organized society to maintain itself.
This is happening sooner than you think.
Modern nation states are a hell of a lot more resilient than what you give them credit for. Look at some of the hardships nations survived during WW2. States can survive even mass famines. How? Simple rule. The police and the army eat first.
For real, Russia is calling with a word for OP.
If it’s any comfort, there’s a good chance the sunlight won’t be reaching the suface of the earth for much of 2027, so we won’t be alive for too long.
All LLM code output is now copyleft because there was GPL stuff in the training data, LOL!
spoiler
(Actually it’s probably all just copyright infringement and not usable at all because of all the conflicting licenses, but a guy can dream…)
I think LLM code output is already copyleft because non-humans legally can not hold copyrights.
Copyleft is a way of leveraging copyright against itself to ensure nobody else can make a proprietary version of the thing. It is not the same as Public Domain/lack of copyright.
in my defense there are a lot of words to remember
LLM output doesn’t automatically get a copyright. But, if it is (part of) a work that includes “human creative effort” (prompts don’t count), the human(s) can hold a copyright on that work.
In addition, the output can still be a derivative work in violation of the copyrights of (some of) the training data, whether or not there are copyrights on that output. It would have to have sufficient similarity to some work in the training data, but that’s not too uncommon.
And, GPL and CC-SA works are known to be in the training data of most models, including Apertus.
:(
