Was AI trained on other people's work without permission?
Books, lawsuits, and what it means for you.
To build a model, companies gather enormous amounts of text: web pages, books, articles, forum posts, code.
Scraping: using software to copy web pages automatically and in huge numbers.
Much of that material is protected by copyright. Whether training on it is allowed is being fought out in court right now.
In August 2024, a group of authors sued Anthropic, saying it had used pirated copies of their books to train Claude.
In June 2025, the judge ruled in two parts:
Fair use: a US legal rule that allows some use of copyrighted work without permission, such as quoting a book in a review.
In September 2025, Anthropic agreed to pay $1.5 billion to settle the case.
A judge gave final approval in July 2026. It was described as the largest known settlement in a US copyright case. Some authors and publishers opted out and are suing separately.
The New York Times is suing OpenAI and Microsoft over the use of its articles. That's the case that led to the order to keep ChatGPT logs (see "Where do my AI chats go?").
Rules differ by country. "Fair use" is American. The UK and EU have their own, different rules, and they're being argued over too.
Regurgitation: when a model reproduces chunks of its training text nearly word for word.
It's rare in normal use, but it happens, especially with famous text, song lyrics or code. If you publish what it wrote, you're the one publishing it.
Don't ask it to reproduce copyrighted text, like a whole article, chapter or set of lyrics.
For anything you'll publish or sell, check your company's AI policy. Many have rules on this.
If something it wrote sounds too polished or too familiar, search for a distinctive phrase from it to make sure it isn't copied.