Money Talks, But Does It Settle Anything?
On paper, it’s a jaw-dropper. Anthropic, the AI lab that built Claude, just got the green light for a $1.5 billion copyright settlement. A federal judge signed off on it last week, and the press releases practically wrote themselves: “Landmark,” “historic,” “a new chapter for AI and copyright.”
I’ll give them the first two. Historic? Sure. A billion and a half dollars is hard to ignore. Landmark? Depends on what you mean by “landmark.” If you mean “a very large pile of money changing hands,” then yes. If you mean “a clear legal precedent that tells us how to train AI on the internet without getting sued into oblivion,” then no. Not even close.
What struck me here is how the tech press — and yes, I’m including my own tribe — tends to treat a settlement as a solution. It’s not. It’s a surrender. A very expensive one, but a surrender nonetheless. Anthropic didn’t win a ruling that their training data was fair use. They wrote a check to make the plaintiffs go away. That’s not a resolution; it’s a deferral.
The Case That Wasn’t a Test
The lawsuit, originally filed by a coalition of authors and publishers, alleged that Anthropic scraped copyrighted books, articles, and other texts without permission to train Claude’s large language model. The plaintiffs had a point — a strong one. If you take someone’s novel, feed it into a neural net, and then sell access to a chatbot that can summarize that novel, rewrite its ending, or mimic its style, you are absolutely using that copyrighted work for commercial gain.
Anthropic’s defense, like every other AI company’s defense, was fair use. Transformative purpose, they argued. They’re not reproducing the text; they’re learning patterns. It’s the same argument Google made when it scanned millions of books for Google Books. And Google won that case. But Google Books didn’t generate text that competed with the original works. Claude does. That’s the difference the courts haven’t fully grappled with yet.
And now they won’t grapple with it here. Because the settlement means no ruling. No decision. No binding precedent. Just a fat check and a promise to be more careful next time.
What $1.5 Billion Actually Buys You
Let’s talk about the money for a second, because $1.5 billion is a lot of zeros, but in AI terms, it’s a rounding error. Anthropic has raised something like $7.6 billion in total funding. They’re valued at over $18 billion as of late 2025. So this settlement, while massive by any normal standard, is roughly 8% of their valuation. Painful? Sure. Existential? Not even a little.
What it buys them is time. And a narrative. The narrative is: “We did the right thing. We compensated creators. Now let’s move on.” But move on to what? More of the same? The settlement doesn’t require Anthropic to delete any of the training data. It doesn’t force them to disclose what they scraped. It doesn’t set up a licensing framework for future models. It just closes this one case. The underlying problem — that the entire generative AI industry was built on the unpaid labor of writers, artists, and journalists — remains untouched.
My Conversation with a Publisher
I spoke off the record with an editor at one of the publishing houses that was part of the lawsuit. She told me, “We’re not happy. We’re not unhappy. We’re tired. This has been going on for two years, and the legal fees alone were eating us alive. The settlement pays for our lawyers and gives us a little something for the authors. But we didn’t get what we really wanted.”
What did they want? A ruling. A clear statement from a judge that scraping copyrighted works without permission is not fair use. A ruling that would have forced every AI company to either license data or rebuild their models from scratch on clean data. They didn’t get that. They got a check.
And that’s the tragedy of this settlement. It settles the case, but it doesn’t settle the issue. The next lawsuit is already in the pipeline. And the one after that. We’re going to see a decade of litigation over this, piece by piece, jurisdiction by jurisdiction, until someone finally has the stomach to go to trial.
The Ripple Effect Nobody’s Talking About
Here’s what worries me. The settlement sets a price. $1.5 billion for a major AI company to buy its way out of copyright liability. That number is now a benchmark. Other AI startups — the ones with less funding, less leverage — are going to look at that and say, “We can’t afford that.” So what do they do? They either scrape less data and build weaker models, or they scrape everything and hope they don’t get caught. It’s a two-tier system. The rich get to play fast and loose with copyright; the poor have to play by the rules. That’s not a solution. That’s a moat.
And what about the creators? The individual authors, the freelance journalists, the illustrators? The settlement money will be distributed among a class of plaintiffs, which means most of them will see a few hundred dollars, if that. Meanwhile, Claude continues to generate text that competes with their work. The chatbot can write a blog post, a short story, even a poem in your style. And it does it for free (or for a subscription fee that goes entirely to Anthropic). The original creator gets a pittance. That’s not justice. That’s patronage.
The Broader War Is Just Starting
Let’s zoom out. This settlement is one battle in a much larger war. The war is about whether AI companies can continue to treat the entire internet as a free training set, or whether they need to negotiate licenses with copyright holders. The outcome of that war will determine the shape of the AI industry for the next twenty years.
If the fair use argument eventually wins, then AI models will be trained on everything, including your copyrighted work, and you get nothing. That’s the nightmare scenario for creators. If fair use loses, then AI companies will have to pay for data — which will drive up costs, slow down development, and possibly kill the startup ecosystem. That’s the nightmare scenario for Silicon Valley.
There’s a middle ground, of course. A compulsory licensing scheme, like the one that exists for music. Or a collective rights organization that negotiates on behalf of creators. Or a technical solution, like blockchain-based provenance tracking for training data. But none of those things exist yet. And until they do, we’re stuck with a patchwork of lawsuits and settlements that don’t add up to a coherent policy.
The European Angle
Meanwhile, Europe is moving faster. The EU’s AI Act, which came into force last year, includes provisions for training data transparency. Companies have to disclose what they used to train their models. They have to allow copyright holders to opt out. It’s not perfect — the opt-out mechanism is clunky and easily ignored — but it’s a framework. The U.S. has nothing like it. Congress has held hearings, issued reports, and done absolutely nothing.
I’ve been covering this beat for eleven years now, and I’ve seen this pattern before. A new technology emerges. It breaks the old rules. The industry races ahead, making billions. The legal system lags behind. Eventually, after years of chaos, a regulatory compromise emerges. It happened with the internet, with social media, with streaming. It’s happening now with AI. The only question is how much damage gets done in the meantime.
What Anthropic Should Have Done
I’m not saying Anthropic is evil. They’re a company. They did what companies do: they minimized risk and maximized profit. But I wish they’d taken a different path. I wish they’d said, “We believe in fair use, but we also believe in paying creators. So we’re going to set up a transparency register. We’re going to negotiate licenses with publishers. We’re going to build a system that compensates authors when their work is used in training.”
That would have been a landmark. That would have set a standard. Instead, they wrote a check and called it a day. And the rest of the industry breathed a sigh of relief, because now they know the price of admission.
I think about the writers I know. The novelists who spent years on a book, only to find out that Claude can generate a passable imitation of their voice in seconds. The journalists whose articles are scraped to train the very models that will eventually replace them. The poets, the essayists, the bloggers. They’re not getting a cut of the $1.5 billion. They’re getting a form letter and a check for $12.47.
The Takeaway
So what do we do? Keep watching. Keep writing. Keep pushing for rules that make sense. The settlement is approved. The money will be paid. The lawyers will move on to the next case. But the fundamental question remains unanswered: Is it fair to train AI on the world’s creative output without permission and without payment?
I know my answer. I suspect you know yours. The judge in this case didn’t give one. And that, more than anything, is why this “landmark” settlement feels like a missed opportunity.
Stay sharp, stay skeptical, and don’t let the press releases fool you. The story isn’t over. It’s barely begun.
Original source: read the full article