AI chatbots like ChatGPT, Gemini, and Claude are built using huge collections of text. That text includes millions of books, articles, and papers pulled from across the internet.
Most authors never agreed to have their work used this way. Many did not even know it happened.
This raises an obvious question. Is it legal to train AI on copyrighted books without permission?
The answer is complicated. Attorney Cathy Gellis, who specializes in intellectual property law, says the topic involves a lot of moving parts and strong opinions on both sides.
What The Anthropic Ruling Actually Said
Last year, Judge William Alsup ordered Anthropic to pay $1.5 billion to a group of authors. The money settled a case over books used to train the company's AI models.
At first glance, this looked like a win for writers. But Alsup actually ruled that training AI on books is lawful.
The fine was not for training on the books. It was for pirating them from illegal shadow libraries instead of buying them legally.
Alsup compared an AI reading text to a writer studying literature. He said the goal was to create something different, not to copy or replace the original work.
Gellis says this outcome favors AI companies overall. She points out that $1.5 billion is a small amount for a company expected to bring in around $200 billion a year by 2028.
Copyright law has not changed since 1976. Judges are left applying decades old rules to brand new AI technology.
Attorney Jason Henderson says this creates confusion across the legal system. He explains that the law simply has not caught up with how much material AI models are trained on.
The Fair Use Question
Much of this debate comes down to fair use. Fair use allows people to use copyrighted material without permission in certain situations, such as criticism or education.
Judges weigh factors like the purpose of the use and its effect on the market. Henderson says courts tend to allow AI training unless it directly competes with the original work.
He points to a separate case involving Thomson Reuters and Ross Intelligence. In that case, a judge ruled it was not fair use because Ross built a competing legal platform using Reuters' content.
Authors could argue that AI generated books compete with their own work. So far, courts have not accepted that argument.
Gellis says it helps to separate two different issues. Training AI on books is one question. Whether AI generated content can be copyrighted is another.
In a separate case, Thaler v. Perlmutter, a court ruled that fully AI generated work cannot be copyrighted. That raises new questions about how much human involvement is needed to claim ownership.
Most AI companies are still facing lawsuits over these training practices. That means a final legal answer will not arrive soon.
Gellis says early rulings are shaping how the industry moves forward right now. She adds that future court decisions could still change the picture as more cases work through the system.
