Judge reluctantly sides with Meta in Sarah Silverman, authors' AI copyright suit

SAN FRANCISCO (CN) — When artificial intelligence can write entire books in seconds, where does that leave the authors of our generation?

A federal judge granted partial summary judgment to Meta on Wednesday, despite claims from 13 award-winning and acclaimed authors who argued that the tech giant violated U.S. copyright law by using their works to train its flagship AI model Llama without their permission.

Although U.S. District Judge Vince Chhabria seemed to sympathize with the plaintiffs, he said he had “no choice” but to rule in Meta’s favor, given the poor arguments by the authors.

However, he said that the public shouldn’t confuse his decision with a blessing on Meta’s behavior.

“It stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one,” the Barack Obama appointee said.

At issue is Meta’s practice of including copyrighted works, like the authors’ books, in the datasets it used to train Llama without paying them. Although Meta’s AI itself is open source, the authors claim the Silicon Valley giant pirated their unlicensed works through “shadow libraries” online, like Library Genesis and Anna’s Archive, and stripped them of relevant copyright information like ISBNs, copyright symbols and disclaimers to conceal their wrongdoing.

“Whatever the merits of generative artificial intelligence, or GenAI, stealing copyrighted works off the Internet for one’s own benefit has always been unlawful,” the authors said in their motion.

The case boasts an impressive roster of plaintiffs. Sarah Silverman, a stand-up comedian and the author of several books, is the most well-known plaintiff in the lawsuit, alongside authors like Junot Díaz, Ta-Nehisi Coates, David Henry Hwang and Laura Lippman, among others.

The authors want declaration that Meta violated U.S. copyright law, as well as damages to be determined at trial.

Meta insists that its use of the authors’ work qualifies as fair use, no matter how they were obtained, and that its use in an AI context counts as a “transformative” use of the information.

“In sum, Meta’s copying of datasets containing plaintiffs’ books was for a transformative fair use and ‘not an infringement of copyright,’” the company said in court filings.

As evidence of copyright infringement, the authors argued that the AI’s responses are sometimes verbatim excerpts from their books, which they claim amounts to Meta unlawfully benefiting from their works by failing to license them through their authors.

They also argued that Meta, by using their works to train AI without their permission, has diminished their ability to license their works to train other large language models in the future.

Chhabria said both of these arguments were “clear losers.”

“Llama is not capable of generating enough text from the plaintiffs’ books to matter, and the plaintiffs are not entitled to the market for licensing their works as AI training data,” Chhabria found.

Chhabria admonished the plaintiffs for barely paying lip service to a potentially winning argument — that Meta copied their works to create a product that will likely flood the market with similar works, thus diluting the demand for their material and harming their book sales.

The judge said that if the authors presented any evidence on the issue, the case would have needed to go to a jury. He even suggested that they could have made a strong enough showing to win on the fair use issue at summary judgment.

But lacking that, the decision belonged to Meta, he ruled.

“That conclusion may be in significant tension with reality, but it’s dictated by the choice the plaintiffs made to put forward two flawed theories of market harm while failing to present meaningful evidence on the effect of training LLMs like Llama with their books on the market for those books,” Chhabria said.

Chhabria also ruled against the authors’ claims that Meta violated the Digital Millennium Copyright Act.

However, the judge said his ruling wasn’t necessarily a huge win for Meta, nor a colossal setback for artists in their struggle against AI-generated content, because it only affects the rights of these thirteen authors — not the countless others whose works Meta used to train its models.

“And, as should now be clear, this ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful,” Chhabria said.

The judge said that in the future, it’s likely that plaintiffs in similar situations as these authors will often win such cases.

“No matter how transformative LLM training may be, it’s hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of a potentially endless stream of competing works that could significantly harm the market for those books,” the judge said.

Chhabira also refuted the idea that ruling against AI companies would stop the development of the technology in general, saying it was “ridiculous” and not a reason to rule in Meta’s favor.

With the authors’ claims effectively whittled down before trial, the judge scheduled a case management conference with both parties on July 11 to address their other claim that Meta illegally distributed their books while torrenting them, which will proceed unaffected by his decision.

The authors’ lawsuit has come a long way over its nearly two-year lifespan. They filed their class action against Meta in July 2023, and the case was initially dismissed in November 2023. However, the judge allowed the plaintiffs a chance to repair their lawsuit.

At a February hearing, the judge criticized the authors for their “over-the-top” rhetoric in the case and warned that if it continued, it could hurt their credibility with the court.

Meta released Llama in February 2023 as an open-source competitor to ChatGPT. The program’s name, originally stylized as “LLaMA” is a near-acronym for “Large Language Model Meta AI,” and it was marketed as a more customizable model that would better suit the needs of software developers and researchers.

Meta said that since its 2023 release, Llama has been downloaded an average of 1 million times per day. The company released its most recent version, Llama 4, in April.

The case was filed in the Northern District of California.