Opinion | Even Millions Of Stolen Books Cannot Satisfy Ravenous A.I. Chatbots – The New York Times

The New York Times has published an opinion item whose headline argues that even millions of books described as stolen cannot satisfy the data appetite of artificial intelligence chatbots. The publication puts renewed focus on copyright and training-data disputes, but the available record contains only the headline, leaving its evidence, targets and legal reasoning unconfirmed.

The item is explicitly labeled opinion, meaning its central proposition should be understood as an argument rather than a reported finding established by the headline alone. Its wording combines two separate assertions: that millions of books were stolen and that even a collection of that scale would be insufficient for chatbot development. Neither assertion can be independently checked without the column’s text, citations or supporting records.

The headline does not identify any AI company, chatbot, book collection, author or acquisition method. It also does not explain whether “stolen” refers to alleged copyright infringement, unauthorized downloading, breach of contract or another legal theory. The available information contains no court ruling, dataset inventory or company response establishing that the books were unlawfully obtained.

The reference to “millions” suggests that the column addresses the scale of material used to build or improve generative AI systems. The description of chatbots as “ravenous” is rhetorical language from an opinion headline, not a technical measurement of how much data a particular model requires. No model version, benchmark or training run is named.

Book Supply Meets AI Demand

The argument matters because disputes over AI training increasingly turn on two different questions: whether developers had permission to use particular works and whether more data produces better systems. The headline links those issues by suggesting that a very large collection may be both legally contested and unable to satisfy developers’ continuing demand.

For authors, publishers and readers, the stakes include control over copyrighted work, possible compensation and the availability of reliable information about model training. For AI companies, the debate concerns the provenance, quality and quantity of training material. The headline offers no evidence that every chatbot requires millions of books, or that books play the same role across different systems.

Copyright Questions Behind the Headline

Books can provide long-form prose, structured arguments and varied subject matter, making them potentially useful within broader AI training collections. Yet the headline does not establish that books were used by a named developer or obtained through a particular dataset. It also gives no information about licenses, public-domain works, purchased copies or material supplied under agreements.

The distinction between an opinion argument and a verified legal finding is central here. Calling material “stolen” may express the columnist’s view of unauthorized use, but courts determine liability from specific facts and applicable law. No judgment, complaint or settlement is identified in the available record, and Anthropic is not named in the headline as the company at issue.

“Even Millions of Stolen Books Cannot Satisfy Ravenous A.I. Chatbots”

— The New York Times opinion headline

Evidence and Targets Remain Unidentified

It is not yet clear which books, datasets or AI systems the opinion item discusses. The available headline provides no author name, publication date, figures beyond “millions,” or links to documentation. It also does not reveal whether the number describes downloaded files, unique titles, copies, training examples or a broader estimate.

The legal status of the material is also unresolved from the available information. No evidence is provided showing who obtained the books, whether permission was sought, how any works were used or whether a court found wrongdoing. The headline’s claim about unsatisfied demand is similarly unquantified, leaving the meaning of “cannot satisfy” open to interpretation.

Full Column and Records Needed

A fuller account will require access to the complete opinion column, including any cited lawsuits, datasets, technical research or company statements. Those materials would show whether the author is describing a documented event, advancing a policy argument or using the scale of book collections as an analogy for AI’s broader demand for data.

Readers should also watch for responses from any companies or rights holders identified in the full article. Court filings, licensing disclosures and dataset documentation could clarify what was acquired, whether use was authorized and how books contributed to a named model. Until then, the headline supports reporting on the publication of an argument, not a finding that millions of books were stolen.

Key Questions

Does the headline prove that millions of books were stolen?

No. “Stolen” is an attributed characterization in an opinion headline. The available record contains no supporting court decision or documentation.

Which AI company is accused?

The headline identifies no company or chatbot. Any claim that a specific developer is the target would be unsupported by the available wording.

Why would AI developers use books?

Books may supply long-form language and structured material for training collections. The headline does not establish how any named model used them or whether permission was obtained.

What information could confirm the opinion’s claims?

Confirmation would require the full column and its evidence, along with relevant court records, dataset inventories, licenses and responses from identified parties. Those records could establish the scale, provenance and legal status of the books at issue.

Source: Anthropic

Source: Anthropic