Who Actually Governs AI: Part 1 - Copyright

In July 2026, a federal judge signed off on a $1.5 billion settlement. Anthropic would pay roughly $3,000 to each of the authors whose books it had trained Claude on, books it pulled from pirate libraries years earlier without asking anyone.

It is the largest copyright settlement in the industry's short history. It is also, against Anthropic's own revenue, a rounding error.

The AI industry never figured out whether training a model on someone else's work without permission is legal. What it figured out, instead, is exactly what that uncertainty costs, and that the cost is one most companies can simply pay.

Two Bills, Two Very Different Ways to Pay Them

Picture two companies staring at the same problem from opposite sides of a courtroom door.

Anthropic trained Claude on hundreds of thousands of books, many of them pulled from LibGen and other pirate libraries. When authors sued, a judge ruled that training on legally acquired books could count as fair use, but training on pirated copies could not. Anthropic settled rather than fight that second question at trial. The bill: $1.5 billion, spread across 482,460 works, with authors claiming roughly 93% of that list.

News Corp took the other route. Five months after the New York Times sued OpenAI and Microsoft over the same underlying issue, News Corp signed a five-year deal with OpenAI reportedly worth more than $250 million. In exchange, OpenAI gets permission to train on and surface content from the Wall Street Journal, the New York Post, and a dozen other News Corp titles. No lawsuit required.

One company paid to make a piracy problem disappear. The other got paid up front so it would never have one.

The Question Nobody Has Actually Answered

Here is the part that makes both of those numbers strange: courts still haven't decided the core legal question.

The New York Times' lawsuit against OpenAI and Microsoft, filed in December 2023, is still active. In September 2026, all three parties filed competing motions asking a judge to rule, once and for all, whether training a language model on copyrighted news articles counts as fair use. A ruling isn't expected before a possible trial next year.

The U.S. Department of Justice weighed in that same month, urging the court to avoid treating AI training as automatically infringing, citing competitiveness with China as a reason to leave room for it.

So the industry has settled a $1.5 billion piracy case and signed dozens of licensing deals, all while the actual fair-use question that would decide whether any of this was ever necessary remains open. Money moved. The law didn't.

Did You Know?

The 482,460 "works" in the Anthropic settlement were identified from pirate libraries called LibGen and PiLiMi, essentially online archives of illegally copied books. Each verified match to the settlement's official list is worth a flat payout, regardless of how many times the book was actually used to train Claude.

Winning in Court Doesn't Always Mean Winning

Getty Images sued Stability AI in both the UK and the US over Stable Diffusion, the image-generating model, and the results show how messy this actually is.

In the UK, Getty lost its core copyright claim. British courts ruled that scraping Getty's images to train the model overseas wasn't infringement under UK law. But Getty won on a narrower point: Stable Diffusion sometimes generated images with Getty's watermark still faintly visible, and that counted as trademark infringement, a completely different legal theory than the one Getty actually cared about.

In the US case, decided in April 2026, a different court let Getty's trademark and unfair-competition claims move forward while tossing a narrower copyright-management claim. Neither ruling settled whether training itself was legal. Both left that question for someone else's case.

When Everyone Settles, Nobody Sets the Rule

This is the pattern showing up across the industry, and it isn't an accident.

A settlement resolves one company's exposure. It does not create a legal precedent other companies can rely on. So every AI company facing a copyright claim has an incentive to settle quietly rather than risk a ruling that could bind the whole industry, and every publisher facing a well-funded AI company has an incentive to take a licensing check rather than gamble years on a trial.

That incentive lines up neatly for both sides, which is exactly why the fair-use question keeps not getting answered. OpenAI alone has now signed roughly 20 verified publisher deals, covering more than 160 outlets, including the Associated Press, the Financial Times, Condé Nast, and Reddit. Anthropic and Google face active suits from publishers and, separately, from Universal Music, Concord, and ABKCO over billions in claimed damages tied to AI-generated songs built on their catalogs.

Some AI licensing deals do more than pay for past use. They also grant the AI company a live feed of new articles, letting a chatbot answer questions about that morning's news the moment it publishes, in exchange for a share of the licensing fee.

The Publishers Big Enough to Get a Seat at the Table

Look at who actually gets a licensing deal, and a pattern shows up fast: it's the publishers who could otherwise afford a lawsuit.

News Corp, the Associated Press, and Condé Nast have the legal budgets and reach to sue and plausibly win, or at least cost an AI company real money and bad headlines trying. That leverage is what got them a check instead of a courtroom date.

Individual authors and smaller publishers rarely have that option. Many only got compensated at all because a class-action settlement, like the one against Anthropic, bundled their claims together automatically, whether or not they knew their book had been used in the first place.

The freelance writers, independent bloggers, and smaller outlets whose work also trained these models mostly got neither a check nor a class action. Their material still shaped the models. Nobody negotiated on their behalf, because nobody was big enough to be worth suing over.

Knowlegic Perspective

The AI copyright fights of 2026 look, from a distance, like an industry cleaning up its past. They are really an industry pricing its future.

Every settlement and every licensing deal is, in practice, a bet: that paying now is cheaper than losing a fair-use ruling later, and that the companies with enough leverage to force a payment will keep getting paid while everyone else absorbs the cost of training without a seat at the table.

That's a workable business strategy. It is not the same thing as an answered legal question, and the gap between the two is where most of the internet's actual writers, the ones without a legal team or a class action to fall into, are quietly left out.

The AI industry didn't resolve whether training on other people's work without asking is legal. It found out what that uncertainty costs, and decided it's a bill worth paying.

Sources & References

• Anthropic to pay authors $1.5 billion over pirated books used to train Claude in first major AI copyright settlement, Fortune (2026)
• What Authors Need to Know About the Anthropic Settlement, The Authors Guild (2026)
• OpenAI and News Corp Strike a Content Deal Valued at Over $250 Million, Nieman Journalism Lab (2024)
• Getty Images (US), Inc. v. Stability AI, Ltd., Loeb & Loeb LLP (2026)
• AI in litigation series: An update on AI copyright cases in 2026, Norton Rose Fulbright (2026)
• OpenAI's Publisher Deals Are Reshaping the News Behind AI Answers, citybiz (2026)