Is it legal to Train AI Models on Copyrighted Books?
Whether tech companies can legally train AI models on copyrighted books has become one of the most explosive and fiercely debated intellectual property battles of the modern era. You probably know by now that the massive large language models (LLMs) powering ChatGPT, Gemini, Claude, and other chatbots are trained on seemingly infinite databases of published works. These datasets contain hundreds of millions of books, online articles, academic papers, and basically anything you can find on the internet.
Most published authors have, without their knowledge or explicit consent, contributed to the development of the exact same artificial intelligence tools that threaten to undermine their livelihoods. That seems highly illegal, right?
The reality isn’t that simple. Just as emerging algorithmic technology continues to trigger unprecedented academic disruptions—such as the unfolding AI crisis in math—the federal legal system is currently struggling to adapt to machine learning.
Here are five shocking legal truths that explain exactly what happens when tech giants train AI models on copyrighted books.
1. The Legal Distinction Between “Reading” and “Copying”
At the core of the debate over whether developers can train AI models on copyrighted books is a fundamental question: Is artificial intelligence “reading” or “copying”?
Last year, in a landmark ruling involving AI startup Anthropic, Judge William Alsup ordered a mammoth $1.5 billion copyright settlement favoring a group of writers. At face value, this seemed like a massive moral victory for authors. However, Judge Alsup actually ruled that Anthropic’s underlying AI training method was lawful. What the judge penalized Anthropic for was pirating these books from illegal online shadow libraries.
“Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different,” the judge wrote.
According to legal analyses highlighted by TechCrunch, copyright law hinges on the act of copying. It does not prohibit consuming, experiencing, reading, or learning from a copyrighted text.
2. The High Stakes of the “Fair Use” Doctrine
In the United States, lawsuits surrounding the right to train AI models on copyrighted books usually hinge on the doctrine of fair use. Codified under the official U.S. Copyright Office regulations, fair use acts as a carve-out of copyright law. It allows the use of protected material without permission for purposes like commentary, education, research, criticism, and parody.
Federal judges evaluate specific factors to determine fair use:
- The purpose and character of the use (Is it “transformative”?).
- The nature of the original copyrighted work.
- The amount and substantiality of the portion taken.
- The economic impact of the use on the potential market value of the original work.
If an AI developer transforms the training material into an entirely new generative capability rather than spitting out a direct, plagiarized copy of a novel, the courts tend to view the action as legally permissible fair use.
3. Direct Market Competition Creates Heavy Liability
While broad algorithmic training may qualify as transformative, building tools designed to directly replace the original creator drastically changes the legal calculation.
A key precedent is the ongoing case where media company Thomson Reuters sued the research firm Ross Intelligence. Ross Intelligence copied proprietary Reuters content to build a competing, AI-based legal platform. The judge determined that using proprietary content to create a platform with the exact same commercial purpose was not fair use.
If a business attempts to train AI models on copyrighted books with the explicit intent to undercut and directly compete with those exact authors in the marketplace, courts are far more likely to issue severe penalties for copyright infringement.
4. AI Training Data vs. Output Copyrightability
When discussing how companies train AI models on copyrighted books, it is helpful to narrow down what we are actually talking about. Courts draw a strict line between the data fed into an algorithm and the content the algorithm produces.
Under the precedent established in Thaler v. Perlmutter, courts ruled that if a work is 100% generated by artificial intelligence, it cannot receive copyright protection. This creates a deeply complex legal landscape: developers can lawfully utilize copyrighted literature to train their foundational systems, but the resulting autonomous outputs may enter the public domain without traditional copyright protections.
5. Applying a 1976 Law to 21st-Century Algorithms
Much of the current legal friction stems from the fact that U.S. copyright statutes have not undergone a comprehensive, foundational update since 1976. Federal judges are tasked with applying 50-year-old frameworks to multi-billion-dollar neural networks.
“Everybody is very worried right now because the law is all over the place,” Jason Henderson, an IP attorney, recently noted.
Most AI companies remain lodged in pending litigation over these issues. Until higher appellate courts or federal lawmakers establish clear statutory guidelines, the decision to train AI models on copyrighted books will remain caught in a legal gray area. Tech firms and creative professionals alike must monitor these early rulings closely, as they will establish the permanent boundaries of the artificial intelligence industry.
Frequently Asked Questions
Can tech companies legally train AI models on copyrighted books?
Under current preliminary rulings, ingesting text to build general predictive language capabilities is often viewed as transformative fair use, provided the training data was obtained through lawful sources rather than pirated shadow libraries.
Why was Anthropic fined if AI training is considered lawful?
Anthropic was penalized for acquiring its training material from unauthorized, illegal pirate repositories, not for the technical act of processing text through its neural network to learn language patterns.
Can authors prevent AI companies from using their literature?
Authors are actively pursuing class-action lawsuits and pushing for new licensing mandates. However, current federal law does not yet provide a standardized, mandatory opt-out mechanism for published print works when companies train AI models on copyrighted books.
