Artists Sue Google, Meta and Anthropic Over AI Training on Their Books

In July 2026, writers and artists launched a series of lawsuits against Google, Meta and Anthropic, accusing the AI companies of training models on pirated copies of their work. The trigger was a database published by The Atlantic, which lets authors search for their works among the texts used for training.

What set off the wave of lawsuits

The push came from an Atlantic publication: the outlet released a searchable database of works included in AI models' training sets. Authors began typing in their names and finding themselves. Nonfiction writer Kirk Wallace Johnson found both of his books there, "The Feather Thief" and "The Fishermen and the Dragon", each of which, by his account, took five to six years of research and writing. In effect, his texts were pirated and handed over to a chatbot.

Why authors are going to court

The authors accuse Google, Meta and Anthropic of using their works to train neural networks without permission or payment. For many writers, these are the products of years of labor that ended up in the training data of commercial models.

• Defendants in the suits: Google, Meta and Anthropic
• What made the issue public: The Atlantic's database, which lets users search training texts
• Example plaintiff: nonfiction writer Kirk Wallace Johnson, author of "The Feather Thief" and "The Fishermen and the Dragon"
• Cost to the author: five to six years of research and writing per book
• According to The Verge, artists are already winning some of these cases

Johnson describes his feelings as a "cocktail" of several emotions.

"Anger at the audacity of this theft, anxiety about what it means for writers, and a healthy thirst for revenge against these giant corporations that have gotten fabulously rich," —

Kirk Wallace Johnson, writer, in The Verge.

What this means for AI companies

The wave of lawsuits moves the dispute over copyright and model training from public debate into the courtroom. Artists and writers used to be able to do little more than vent on social media; now they have a verification tool, The Atlantic's database, that makes it easy to show a specific work ended up in the training data. As The Verge notes, authors are already taking some cases to victory, which means the companies' appeal to "fair use" is no longer a guaranteed defense.

What it means

The conflict between authors and the AI industry is entering a stage where claims are backed by evidence and court rulings, not just statements. For Google, Meta and Anthropic, that means growing legal and financial risk over the data their models were trained on.

*Meta is designated an extremist organization and banned in Russia.