Book publishers sue Meta over alleged copyright use in Llama AI training

In short: A group of major book publishers and one author sued Meta, saying it copied copyrighted books and articles without permission to train its Llama AI models.

What happened

Five major publishers, Macmillan, McGraw-Hill, Elsevier, Hachette, and Cengage, along with author Scott Turow, filed a class action lawsuit against Meta. They claim Meta repeatedly copied their books and journal articles without permission.

The lawsuit says Meta pulled material from well-known pirate sites, including LibGen, Anna’s Archive, Sci-Hub, Sci-Mag, and others. The publishers argue Meta then used that material to train its Llama models. Training an AI model is like giving it a huge library to learn patterns from, but the lawsuit argues this “library” included unauthorized copies.

The complaint also points to Common Crawl, a large public dataset made from web pages, and says it is “full of unauthorized copies of copyrighted works.” The publishers say Llama can sometimes reproduce copyrighted text word for word or almost word for word. The lawsuit gives an example involving a Cengage calculus textbook, where it says Llama continued a section after being prompted with two sentences.

This case follows earlier lawsuits from authors against Meta over similar copyright claims. The Verge notes that a federal judge previously ruled for Meta in one case, but also said that decision did not mean Meta’s use of copyrighted materials for training is always legal.

Why it matters

This lawsuit is part of a bigger fight over who gets paid when AI systems learn from books, news, and other writing. If publishers win, it could push AI companies to license more content, meaning deals and costs that may affect what AI products can do and how much they cost.

Source: The Verge AI