The scientific publishing giant Elsevier has joined a class action lawsuit against Meta Platforms over the tech company’s alleged illicit use of Elsevier books and journal articles to train its artificial intelligence models.
Elsevier, along with the publishers Cengage Learning, Hachette, Macmillan Publishers, and McGraw Hill, as well as the novelist Scott Turow and his company S.C.R.I.B.E., filed the legal complaint in a New York City court last week. The plaintiffs claim that Meta infringed copyright by using millions of copyrighted works to train its large language models (LLMs).
Meta officials have said they will use fair use—a US doctrine that allows limited usage of copyrighted works without requiring permission from rightsholders—as a legal defense.
Last October, in a first-of-its-kind settlement, the AI company Anthropic agreed to pay $1.5 billion to authors whose books the firm had used to train its LLMs without first seeking permission.
In the last few years, some scholarly publishers have signed licensing deals with AI companies allowing them to train LLMs on academic papers and data. But Elsevier seems to be taking a different tack, and other scholarly publishers are keeping an eye on how the case progresses.
A spokesperson for the American Chemical Society—which previously accompanied Elsevier in taking the social networking site ResearchGate to court for copyright infringement—says ACS is watching this and other AI litigation closely. (C&EN is published by ACS.)
A spokesperson for the publisher Wiley declines to comment on a lawsuit involving other parties but says that “the use of copyrighted works for AI development should be lawful, transparent, and based on appropriate permissions and licensing.”
The spokesperson adds that Wiley is “committed to protecting the rights of authors, researchers, societies, and copyright holders, while also exploring responsible ways for AI to advance research, education, and knowledge discovery.”
Lisa Janicke Hinchliffe, a librarian and information scientist at the University of Illinois Urbana-Champaign, says it’s unlikely that the issue of how copyright works within generative AI frameworks will be settled anytime soon.
“I don’t think it’s going to be a simple ruling, because it’s a very complex set of accusations,” Hinchliffe says. What’s clear is that this lawsuit will change the risk analyses that AI companies carry out when they seek content and will likely drive them toward more licensing, she says.
But Hinchliffe notes that the risk assessment will also account for the fact that the courts are finding that fair use can be defended, at least in some cases.
In the next stage, the court will determine whether Elsevier and the other plaintiffs can represent a wider class of authors, Hinchliffe explains. If the class is certified, the named plaintiffs bear the legal costs, and anyone in that class is able to then remove themselves if they wish; if it’s not certified, the court may dismiss the case and ask the plaintiffs to instead file complaints individually.
Dave Hansen, executive director of the Author’s Alliance, a US-based nonprofit that supports authors in disseminating their work, notes in a blog post that the lawsuit covers only works that Meta may have used “without authorization.”
What’s more, some open-access journals publish content under various creative copyright licenses that may allow readers to copy, distribute, or even reuse content as long as it is properly attributed. It’s unclear whether these copyright licenses allow AI companies to train LLMs, and this will need to be determined in the courts.
According to Hansen, the core exclusive rights that are the subject of this lawsuit are with Elsevier, not the authors. “An author who specifically pushed to preserve public access to their own work would therefore have no standing to opt out of litigation pursued in the publisher’s name, and no direct say in any settlement that results,” he writes.
