The legal question hanging over every major AI lab right now comes down to one distinction: is training an AI model on a book more like reading it or copying it? That framing is starting to decide cases, and according to TechCrunch AI, the courts landing on “reading” are handing AI companies a quiet but significant advantage.
This matters because the models behind ChatGPT, Gemini, and Claude were trained on hundreds of millions of books, articles, and papers, most of them scraped without the authors’ knowledge or consent. What stands out is how the biggest ruling so far actually cut in the AI industry’s favor, even though the headline number looked brutal.
The $1.5 billion ruling that wasn’t what it seemed
Last year, Judge William Alsup ordered Anthropic to pay a $1.5 billion settlement to a group of writers. At first glance, a win for authors. Read closer, and it’s the opposite.
Alsup ruled that the AI training itself was lawful. What he penalized Anthropic for was pirating the books from illegal shadow libraries, not the act of learning from them. He compared an LLM ingesting trillions of words to a writer studying literature, training “not to race ahead and replicate or supplant them, but to turn a hard corner and create something different.”
Attorney Cathy Gellis, an intellectual property specialist, told TechCrunch the ruling favors AI companies. Her reasoning is blunt: “Copyright law hinges on copying, but it doesn’t hinge on using the work or experiencing the work, consuming the work, reading the work.” And a $1.5 billion fine barely registers against a company projecting roughly $200 billion in annual revenue by 2028.
Why the law can’t keep up
Here’s the structural problem. Copyright law hasn’t been updated since 1976. Judges are interpreting 50-year-old rules to settle disputes that could shape the entire AI economy.
“Everybody is very worried right now because the law is all over the place,” said Jason Henderson, an IP and media attorney at JWL International. The cases keep circling back to fair use, specifically whether a use is “transformative” enough to be permitted without asking. Courts weigh the purpose of the use, how much was taken, and the effect on the market.
Henderson sees a pattern forming. Train on someone’s work to build a product that directly competes with them, and courts push back. Train on it for something that doesn’t compete, and they tend to find room to allow it.
The cautionary example: Thomson Reuters sued Ross Intelligence for copying its content to build a rival legal research platform. Judge Stephanos Bibas ruled it wasn’t fair use, because Ross’s use had no “further purpose or different character” than Reuters’ own. Authors could argue chatbots compete with them by generating synthetic books, but that argument hasn’t won yet.
The other half of the problem
There’s a second question that’s easy to conflate with the first. Training on copyrighted work is one issue. Copyrighting AI-generated output is another entirely.
In Thaler v. Perlmutter, the court ruled that a fully AI-generated work can’t be copyrighted at all. That opens messy follow-up questions: how do you prove something was AI-generated, and how much AI assistance is too much? Gellis put it plainly. If you write a novel in Word and run spell check, nobody thinks Word owns your book. AI is forcing courts to define lines everyone was happy to ignore.
What practitioners should take from this
The fight is far from settled. Most AI companies are still stuck in pending litigation, and early rulings could be reversed on appeal. But the current signals are worth acting on now:
- Sourcing is the real liability. Alsup blessed the training and punished the piracy. How you acquire data may matter more than what you do with it.
- Competition is the tripwire. Building a tool that directly displaces the rights holder you trained on is the fastest way to lose in court.
- Don’t count on owning pure AI output. Fully machine-generated work sits outside copyright protection today.
As Gellis noted, these opening rulings are already shaping behavior, and “it would be kind of foolish for the AI companies to ignore them.” For the full breakdown of the cases and legal reasoning, the original TechCrunch AI report is worth your time.