This case pits real companies and an author against Google in ongoing litigation. Nothing has been decided to date. Here we present the allegations as they appear in the complaint, the arguments Google can legitimately raise, and the precedents that will weigh on the outcome. The alleged facts remain allegations until a court rules.
Some stories come back around to close their own loop. In mid-July 2026, four plaintiffs filed a complaint of around sixty pages against Google before the federal district court for the Southern District of New York. Among them is a bestselling author named Scott Turow. The same Scott Turow who, some twenty years ago, chaired the Authors Guild when it was already taking on Google for digitising millions of books. He had lost. He is back, on the same ground, for a different reason.
Who is suing, and over what
The plaintiffs form an impressive lineup: Hachette Book Group, one of the world's largest publishers, Elsevier, publisher of leading scientific journals such as The Lancet and Cell, Cengage Learning, a heavyweight in educational publishing, and author Scott Turow.
Their accusation: Google allegedly built its Gemini models by copying millions of protected books and articles, without authorisation or compensation, drawing on several sources including Google Books, online libraries and, according to the complaint, pirated content sites. The complaint's text pulls no punches, describing "one of the most prolific infringements in history", and accusing Google of turning its back on its former motto.
The plaintiffs advance two main arguments. First, Google allegedly circumvented a licensing market that was taking shape, in which other AI companies now pay publishers to use their content. Second, and perhaps the strongest point, Gemini would directly compete with the works it was trained on, producing detailed summaries and textbook-style explanations that could reduce demand for the originals. They are seeking damages and an injunction to halt the contested use.
Why the first battle was lost
To grasp the stakes, one must go back to the earlier case. In 2005, the Authors Guild sued Google for mass-digitising books to feed Google Books. After a decade of litigation, the Second Circuit Court of Appeals ruled in Google's favour in 2015, finding that this digitisation fell under fair use. The Supreme Court then declined to review the case.
The judges' reasoning rested on a central point: the public benefit. Google Books allowed users to search within books and display short excerpts, without replacing them. The digitisation was deemed transformative, and it did not substitute for buying the book. Turow had argued the opposite at the time, stressing that Google had copied every word of protected works to enrich its own engine, not out of philanthropy.
The legal distinction is subtle but decisive. In 2015, the use deemed fair was enabling search within books. In 2026, the contested use is training a system that produces competing text. A search engine that shows you three lines from a book makes you want to buy it. An AI that gives you a full summary and educational explanations can, instead, spare you from buying it. It is precisely on this shift that the plaintiffs build their case.
The argument that could tip everything
One element of the complaint deserves particular attention: the allegation that some of the works came from pirated sources. A legal specialist consulted on this case explains why this point is strategic. The fair use defence concerns the use made of a work. But if the initial acquisition of that work was itself illegal, the fair use argument becomes far more fragile, if not inoperative.
In other words, the plaintiffs are not merely attacking what Google did with the books; they are attacking how Google allegedly obtained them. This is a strategy seen in several similar ongoing proceedings against other labs.
Precedents that cut both ways
It would be unwise to predict the outcome, as recent case law is contradictory. In a 2025 case pitting a group of authors against Meta, a federal judge found that AI training on books met the fair use criteria. Conversely, in the proceedings against OpenAI, brought notably by the Authors Guild and writer George R. R. Martin, a judge refused in October to grant the company's motion to dismiss, allowing the case to proceed.
These cases broadly follow the same pattern: protected works were allegedly copied for training, and those copies would be unlawful. But judges are not answering the same way, leaving the law in a state of uncomfortable uncertainty for everyone, creators and tech companies alike.
What is really at stake
Behind this trial lies a question we had already encountered in the distillation case between Anthropic and Alibaba, in another form: what does it mean to own something that does not disappear when copied? A book read by an AI is not stolen from its author. But if the AI draws from it the ability to produce a substitute, something has indeed been transferred, and the question of compensation becomes legitimate.
It is also a debate over the balance between two equally defensible interests. On one hand, creation needs to be remunerated to keep existing, and authors and publishers can hardly fund their work if their output feeds competing products for free. On the other, access to knowledge and technological innovation also hold value for society, and it is exactly that reasoning that tipped the scales in 2015. We had explored this tension from a creative angle in our article on art in the age of AI.
Twenty years on, the same man is suing the same company over the same books. But the technology has changed in nature, and with it the question put to the judges. In 2015, it was about whether you could index a library. In 2026, it is about whether you can turn it into a writer.