Marcel HeadlamShould AI Pay the Writers It Learned From?
Anthropic, the company that makes Claude, paid $1.5 billion to settle a lawsuit brought by authors. Settling means agreeing to pay to end the case. The settlement covers more than 480,000 works. A court gave the settlement final approval on 20 July 2026 (Bloomberg Law and eWeek).
Every smooth paragraph AI hands you was learned from someone's writing. Was that writing stolen?
The slogan says so, but what the court found is more awkward, and worth having an opinion on.
First, I should be open with you before I take a side. I use Claude every day, and the case was against the company that makes Claude.
The case that it's theft
Start with John Locke, whose ideas about natural rights run through the US Declaration of Independence (Monticello). Locke asked how a thing becomes yours.
His answer is in his Second Treatise, and it starts with a person taking something from what everyone shares. That person "hath mixed his labour with, and joined to it something that is his own, and thereby makes it his property" (ch. 5, §27).
What Locke means. If you take something nobody owns and put your own work into it, the result becomes yours. Your effort is what makes it your property.
Why it matters here. A book is the plainest example, because a book is years of a writer's work mixed with a language everyone shares. So by Locke's test, the book belongs to the writer.
And Locke adds a condition, that the claim holds "at least where there is enough, and as good, left in common for others."
What Locke means. You can only take something for yourself if you leave enough, of just as good quality, for everyone else.
Why it matters here. A writer can fairly ask whether that's true of AI, now that a machine produces endless decent writing. Does that leave enough, and as good, for the people whose work taught the machine?
Part of the theft charge stuck when a group of authors sued Anthropic in the United States.
In 2025, the judge, William Alsup, ruled on one part of the case. Anthropic had kept a central library of pirated books, meaning illegally copied ones, and the judge said keeping that library was not "fair use". Fair use is a US legal rule that allows some use of other people's work without asking.
The case that it's learning
The same judge split the question in two, and he found the other half was fair use. That other half was training AI on books, which means letting AI learn from reading them. Locke's labour test cuts both ways here, because an AI company can say it mixed enormous labour of its own with what it read.
There's a simpler point, though, and it's the strongest. No human writer pays every author they've ever read, and if reading created a debt, every novelist would owe thousands.
Learning, with a share owed
The court has said what's legal, but what's fair?
That's your call, and here's mine. Learning from books is fair, because nobody writes without having read. But the people whose work built these tools deserve a say, and a share.
The novelist has a real point, and scale is my only answer to that point. One reader learning from a shelf is not the same as one system learning from hundreds of thousands of books at once. Whether scale changes the principle is the open question.
I'm not a lawyer and this isn't legal advice. This was one US judge in one local federal court, in one case. The case was settled before any appeal, which is when a higher court looks again. So the ruling doesn't have to be followed across the US.
The New York Times' case against OpenAI and Microsoft was still waiting for a decision in a New York federal court in June 2026. Other cases and other countries may come out differently.
Locke was writing about land and acorns in 1690, and he's also a complicated figure, with a troubling record on slavery and colonies. Using Locke here is an argument, and it settles nothing.
What you can do in your own work
Does using AI make you part of the dispute? No, because you didn't train the AI.
Two habits keep your own conscience clear.
First, when AI gives you something you'll put your name to, ask whose ideas it stands on. Adapt this prompt, then check the links, because AI can attach a real name to something that person never wrote.
Whose ideas is this answer built on? Name the writers or researchers
most associated with them. Search the web and link each one so I can check
they exist and said this. If you're not sure, say so.
If an idea comes from a writer, name them, and if their book would help, buy it or borrow it. That's the share, at the scale you control.
Second, ask AI for your own voice instead of a living writer's.
Here are three things I've written. [paste them]
Describe how I write. Then ask me what I'd change about that description
before you draft anything in it.
If you write, and wonder what happens to your words, start with What "Training on Your Data" Means.
Which writer taught you most about how to write, and what do you think you owe them?