Anthropic just agreed to pay $1.5 billion to settle a lawsuit over pirated books.
The ruling that led here was surprisingly generous: training AI models on copyrighted material is fair use, so long as the company acquires that material legally. In other words, buy the book, train away; steal the book, write a billion-dollar check.
But “legal” and “ethical” aren’t synonyms. And that’s where the trouble begins.
Who Gets Paid, and Who Doesn’t
The settlement does include payouts for a handful of authors named as part of the suit. That’s not nothing, and it deserves to be acknowledged. But it doesn’t change the bigger problem: the people who create the content that AI models are trained on are still sidelined in favor of the publishers.
Yes, publishers package and distribute content, but without the researchers, journalists, novelists, and translators doing the actual work, publishers would have nothing to sell. And here’s the reality: most creators were never adequately compensated when the work was first published. Scholars like me spend years producing articles and essays for which we earn precisely nothing. A monograph? If I publish one, I’d be lucky to make enough in royalties to buy a moderately decent bottle of wine. Meanwhile, the publisher sells that book back to my own library at an eye-watering price.
So when publishers cash in on licensing deals with AI companies, it isn’t justice, it’s déjà vu. The original creators are exploited once to produce the work, and again when their work is repurposed for machine training while the publishers pocket the payment. Calling that “fair use” may be legally sound, but it’s ethically anemic.
Anthropic’s Own Goal
Then there’s Anthropic itself. Unlike grad students scrounging PDFs from LibGen at 3 a.m., Anthropic is a multibillion-dollar corporation that markets itself as the ethical AI company. Their “constitutional AI” framework has been hailed as a model for responsible development.
And yet, when faced with the dull mechanics of licensing content, Anthropic went rummaging through pirate libraries. The judge called it theft, and for once, he wasn’t exaggerating.
This isn’t just a legal misstep, it’s a branding disaster. You can’t posture as the conscience of the AI industry while quietly raiding shadow libraries. The settlement doesn’t erase their technical achievements, but it does plant a question mark over their credibility.
The Bigger Picture
The lesson here is twofold. First: AI training can be legally fair use, but until creators—not just publishers—are compensated, it won’t be ethically fair. Second: the AI companies claiming the moral high ground need to prove it with their practices, not just their marketing.
Until both industries (the parasitic publishers and the corner-cutting platforms) are forced to reckon with how they exploit creators, we’ll keep getting rulings that make the gatekeepers richer while the rest of us keep downloading PDFs like contraband, and saving up the guilt for our next trip to the confessional, where naturally it will be the only thing we can afford to give.