Textbook authors sue OpenAI and Microsoft over copyright infringement

The lawsuit against AI giants

A Boston College finance professor, Alan Marcus, joined five other current and former professors to file a proposed class-action lawsuit against OpenAI and Microsoft. The legal action, filed in the U.S. District Court for the Southern District of New York, targets the companies for copyright infringement. The authors claim their copyrighted textbooks were used without permission to train large language models (LLMs) like ChatGPT.

The plaintiffs argue that the companies downloaded their books from "shadow libraries," which are online repositories of pirated content. Specifically, the complaint points to databases known as Books1 and Books2, which allegedly contain ebook files from Library Genesis. The authors estimate that textbooks make up roughly a quarter of all the works the companies copied from these sources.

Microsoft is named in the suit because it provided the infrastructure for OpenAI's training. The complaint alleges that Microsoft's Azure data centers in New York facilitated the exploitation of the training dataset. Furthermore, the suit claims Microsoft profited by integrating these models into commercial products like Bing Chat and Copilot.

Data scraping and concealment

The textbook authors allege that OpenAI went beyond simple downloading by using specific software to process the pirated data. According to the complaint, this software stripped away copyright-management information (CMI), including titles, ISBNs, and copyright notices. The professors argue this was done to "conceal" the use of copyrighted works and to facilitate the training process.

The lawsuit also claims that OpenAI intentionally designed its models to avoid admitting they were trained on pirated or illegal material. The plaintiffs state that OpenAI personnel introduced curated fine-tuning samples to suppress such answers. This was allegedly done to reduce the risk that the public would discover the origin of the training data used for the GPT models.

The scale of this alleged copying is described by the authors as massive. They point to the Books2 dataset, which contains 55 billion "tokens." Tokens are the basic units of text, such as words, syllables, numbers, and punctuation marks, which the professors say suggest the dataset contains over 100,000 books.

Impact on the textbook market

The plaintiffs argue that AI tools pose a specific threat to the textbook industry because of how these books are purchased. In many cases, an instructor or a school district chooses the required text, but the financial burden falls on students. The authors claim that if an AI can provide a "good enough" substitute, it creates a powerful incentive for buyers to skip the original work.

The suit highlights how the models can mimic the specific styles of authors. For example, when asked about a precalculus textbook by Professor Michael Sullivan, ChatGPT reportedly replied that it would use the "same sort of progression" he uses. The authors argue this kind of imitation directly threatens their ability to sell their work to students and institutions.

The group of plaintiffs includes several prominent academics from across the country. These include professors of finance, biology, business communication, and anatomy. They are seeking to represent every textbook author whose copyrighted work was allegedly copied by OpenAI or Microsoft during the development of their models.

The legal battle ahead

The lawsuit was temporarily grouped with other high-profile copyright cases, including the one filed by the New York Times. While the authors have not yet received a full response from the AI companies, the legal landscape is shifting. In early September, a judge stayed the textbook authors' lawsuit until a ruling is made on summary judgment motions in the grouped cases.

Microsoft and OpenAI have already begun defending their practices in other proceedings. Microsoft argued that copyright law should not be used to lock in business models or create monopolies over facts. OpenAI echoed this, stating that "nobody owns facts, just as nobody owns language, grammar, or style." They contend that making facts accessible through AI is a form of progress that copyright law should not hinder.

The Department of Justice (DOJ) recently entered the fray by filing a brief supporting OpenAI. The government argued that training AI models on copyrighted writing is a "fair use," a legal doctrine that allows some copying without permission. The DOJ stated that it is in the national security interest of the United States to allow AI companies to develop these models without significant legal restraints.