Yesterday, a federal judge signed off on the largest copyright settlement in history. Anthropic — the company behind Claude — will pay $1.5 billion to authors whose books were used to train its models. Roughly $3,000 per book, across nearly half a million books.
Most entrepreneurs will read that headline, think "big tech problem," and scroll on.
That's a mistake. Because buried inside this case is the first real answer to a question every business owner should have been asking for two years: how is the law actually going to treat what AI companies did with everyone's content?
Now we know. And the answer cuts both ways.
What actually happened
Two rulings came out of this case, and the distinction between them matters more than the number.
First: Judge William Alsup ruled that training an AI on copyrighted books is fair use. Read that again. The core activity — feeding the world's writing into a model so it can learn — is legal. The tech companies won that argument.
Second: Anthropic kept a library of seven million pirated books, downloaded from sites like Library Genesis. The judge ruled that keeping that library was infringement. Not the training. The hoarding.
Anthropic disclosed the pirated library themselves, in their own discovery documents. They handed the plaintiffs the case.
A company built on machines that never forget was undone by paperwork it couldn't make disappear.
So the settlement isn't a punishment for training AI on books. It's the bill for how the books were acquired. The training itself walked free.
Why this is the landmark
This is the first time a monetary value has been put on the raw material of AI. Not a theoretical value. A court-approved, class-action, cheque-clearing value: about $3,000 a book.
That number is now a reference point. Every AI lab, every publisher, every rights holder, every lawyer drafting the next lawsuit — they're all recalibrating around it today.
Think about what's been established in one case: scraping and training? Fair use. Permitted. The tech companies have effectively been told the door is open. Taking content without paying for it? That now has a price tag, and the price tag has nine zeros.
That's not a loss for either side. That's a tie. The IP owners got paid. The tech companies got their business model blessed. Both walked away with the thing they needed most.
And ties get formalised.
The RIAA model is coming
Here's my prediction, and I'll give it six months to a year.
We end up with the radio model. When a radio station plays a song, the artist gets a set fee. Nobody negotiates per spin. Nobody sues per broadcast. There's a standing rate, a collection body, and a pipe that moves money from the people using the content to the people who made it.
The RIAA and its equivalents built that machinery decades ago, and it's the obvious template here. The Anthropic settlement just wrote the first line of the rate card: $3,000 per book, when acquired badly. The next negotiation starts from there — what does content cost when acquired properly?
Watch for it: licensing bodies for training data. Standing rates for books, music catalogues, news archives, image libraries. Deals signed quietly between labs and publishers who'd rather have recurring revenue than a decade of litigation.
The lawsuits were never going to decide whether AI gets trained on human work. That ship sailed. The lawsuits were always going to decide the price. And now we have the opening bid.
What this means for your IP
This is where it stops being a spectator sport.
Everything you publish — your frameworks, your course content, your newsletters, your podcast transcripts, this blog post — is training data. The courts have now said the training is fair use. You cannot opt your public content out of the machine. That battle is over.
What you can do is get positioned for the world that's coming.
If the RIAA model arrives, your content becomes a licensable asset class. The entrepreneurs who treated their IP like a filing cabinet will get nothing. The ones who treated it like a catalogue — registered, organised, provably theirs — will be holding something a collection system can actually pay out on.
That means owning your copyrights cleanly. Knowing what you've created and where it lives. Registering the work that matters. Keeping the provenance trail tidy — because if this case proved anything, it's that the records decide the outcome. Anthropic's own records cost them $1.5 billion.
And it means thinking hard about the split in your own strategy: what you publish freely because visibility compounds, and what you hold back because it may soon have a licensing value nobody's paying you yet.
Your ideas are now an input to the most valuable machines ever built. For two years, the machines took them for free. Yesterday, for the first time, a court made them pay.
The meter is running. Make sure it's running in your name.
Cheers,
Pete