Justice rules in favor of Meta: its AI may use copyright protected works for training purposes

Meta prevailed. In a ruling issued by Federal Judge Vince Chhabria, the company led by Mark Zuckerberg was exonerated from the accusations attributing to it the unauthorized use of protected works to train its generative AI models, such as the Llama model. This case highlights the tensions between content creators’ rights and technology companies, as the implications of using large volumes of data to train AI algorithms are explored.

The origin of the conflict: the copyright infringement lawsuit

The lawsuit in question, Kadrey v. Meta Platforms, Inc., involves a group of thirteen authors who, in 2023, formally filed a complaint against Meta for copyright infringement. According to the plaintiffs, among whom are authors such as Sarah Silverman, Richard Kadrey, and Ta-Nehisi Coates, the technology company used their copyrighted works without permission to train its artificial intelligence (AI) model, LLaMA. The lawsuit focuses on two main aspects: the elimination of Copyright Management Information (CMI) and the alleged copyright infringement.

For the former, they assert that Meta intentionally removed the copyright management information from their works, which facilitated copyright infringement by allowing Meta’s AI model to be trained with these materials without proper attribution.

The second aspect includes an accusation under the California Digital Crimes Act (CDAFA), alleging that Meta accessed the authors’ books without authorization. However, this latter claim was dismissed. The judge determined that the lawsuit was based on rights already covered by copyright law and that there was no additional valid claim under the CDAFA.

Meta’s response

Meta, through its attorneys, responded to the accusations by alleging that the use of these data was covered under the principles of “fair use” as established by United States copyright law. The company argued that its AI training practices did not infringe copyright because the content was used for research and technological development purposes, rather than for direct commercial gain.

According to this defense, the use of excerpts from books to train AI algorithms fell within this framework, as it was considered that the process did not harm the book market nor affect the value of the works in the marketplace. The technology company argued that, in fact, the use of such content could contribute to the creation of new products and services that would benefit users, such as the improvement of content recommendation systems and the personalization of advertisements.

What the ruling in favor of Meta states

The case moved through several courts and took on greater significance last March, when a United States district judge allowed the lawsuit to proceed, dismissing Meta’s request for its annulment.

Today, the conflict is once again in the news following the ruling issued by Judge Vince Chhabria of the United States District Court for the Northern District of California, which favors Meta. In rendering his verdict, the judge focused on four factors:

1. The purpose and character of the use: The judge considered that Meta’s use of the books was highly transformative. Meta’s purpose was to train a generative AI model to perform tasks very different from the original creation of the literary works. Although Meta prevailed in this regard due to the transformative nature of its use, the judge also emphasized that the commercial use of these models affected the fair use analysis, since Meta anticipated receiving millions of euros in revenue from this technology.
2. The nature of the copyrighted works: In this case, the judge concluded that the works of the plaintiff authors—such as novels, memoirs, and works of fiction—enjoy stronger protection under copyright law due to their creative nature. However, this factor did not have a decisive impact as the amount of material used and the type of transformative use were more pertinent to the ruling.
3. The amount and substantiality of the material copied: Meta copied the works in their entirety for the purpose of training its AI model, but the judge argued that this was reasonable given the transformative purpose of the use. In summary, the volume copied was deemed necessary for the proper training of the model, which justified the amount used.
4. The effect on the potential market for the works: This factor was the most crucial. The plaintiffs argued that the unauthorized use of their books to train the model directly affected the market for their works and the possibility of licensing their books for AI training. However, the judge concluded that Meta had provided compelling evidence that its use did not affect the market for the plaintiffs’ books, and, on the contrary, that generative AI models do not directly compete with the authors’ original works.

The ruling also addressed other legal aspects, such as the downloading of works from “shadow” libraries and unauthorized distribution, but the main focus was the lack of empirical evidence from the plaintiffs demonstrating that the use of their books in AI model training actually impacted the market for their works.

The consequences of the ruling and its impact on the industry

This type of decision highlights how technology can challenge the traditional framework of copyright protection by allowing technology companies to use vast amounts of copyrighted material without having to obtain explicit licenses from the authors.

Could this reduce the commercial value of authors’ works, dilute the market for literary and artistic creations, or open the door to new licensing and regulatory models? We do not know. However, this precedent is certainly not encouraging news for genuine authors and content creators, who once again have witnessed the balance shift toward those in positions of power.