Anthropic Record AI Copyright Pact Sets Bar for OpenAI, Meta (1)

Anthropic’s landmark agreement to pay authors a combined $1.5 billion for downloading pirated book libraries sets a floor for artificial intelligence giants to resolve similar copyright lawsuits.

“This is very sobering for other AI companies,” said Chad Hummel, an attorney at McKool Smith who’s not involved in the litigation. “The content-licensing market will accelerate, and the dollars will be bigger” now that the settlement has been reached. OpenAI Inc. and Meta Platforms Inc. are involved in similar litigation.

Anthropic PBC last week agreed to pay roughly $3,000 for each of nearly 500,000 registered books the authors showed were in pirate libraries the company downloaded and said were maintained in a “research” library of millions of books. California federal Judge William Alsup in August said Anthropic had “refused to come clean” about which pirated works it used to train its Claude model and why it retained them all.

Alsup on Monday blasted the parties at a hearing in San Francisco, rejecting—without prejudice—the proposed settlement as “nowhere close to complete” and expressing concern class lawyers are striking a deal behind the scenes that will be forced “down the throat of authors.”

The agreement—believed to be one of the largest in US copyright litigation history—is among the first to be reached in a litany of disputes in which artists and content owners accuse AI companies of illegally snatching up troves of data—books, movies, song lyrics, or news articles—then feeding that data to build generative AI tools that have created multibillion-dollar valuations. Now those rival AI firms must decide whether to follow Anthropic’s lead or continue fighting legal claims over the seemingly ubiquitous use of the same shadow libraries.

OpenAI is confronting a complex legal challenge defending against a dozen cases from news outlets, including the New York Times, over infringement and trademark dilution. Meta and Midjourney Inc. face their own copyright claims, while music publishers have filed a separate suit against Anthropic over its alleged wholesale copying of lyrics.

The agreement “makes it more likely that authors have rights against OpenAI, Meta, and others” because they’ve admitted to downloading books from the same pirate libraries Anthropic used, Manatt, Phelps & Phillips, LLP partner Tod Cohen said.

A Watershed

The settlement marks the latest watershed moment to force AI companies to evaluate their approach to content acquisition and use of copyrighted content. Earlier this year, Anthropic convinced Alsup that training AI on legally-obtained material qualifies as fair use. A trial was set for December on the authors’ claims related to using Bit Torrent to download pirate libraries.

“It’s definitely started to put guardrails around what can or can’t be done,” said intellectual property attorney Vivek Jayaram.

The deal’s structure—namely the $3,000 per book—establishes a potential benchmark for valuing that content.

“A price tag has been set,” said Peter Henderson, a professor at Princeton University. "$2,000 to $3,000 a book is a recurring theme across the contracting space, across the settlement.”

One challenge in adopting this benchmark is its origins. While the parties may disclose the source of the valuation in settlement hearings, “we don’t know how they arrived at it,” Hummel said. The starting price for literary works may be higher than that for other types of content, he said, such as news articles, which some AI companies argue cover non-copyrightable facts.

The settlement is also limited in scope, covering only “past claims” while separately excluding claims of reproduction, distribution, or the creation of “derivative works” through LLM outputs. The deal would also only cover instances of infringement predating Aug. 25, when the deal was finalized.

It makes sense the authors didn’t release claims regarding outputs, Desai said, because they’re “an evolving and moving target,” and courts haven’t yet weighed on copyright liability for generative AI outputs.

It also requires Anthropic to destroy the “original files” it obtained from the pirate libraries LibGen or PiLiMi, and any copies that originated from the pilfered files, within 30 days of final judgment.

Cohen noted the pact doesn’t require Anthropic to “break” its models and called it a win for the company.

The deal also serves as both a warning and potential roadmap for companies that actually used pirated material in training and is part of the output, a distinction that wasn’t definitively established in Anthropic’s case. Those companies could face even greater exposure if cases go to juries, said Hummel.

Music publishers suing Anthropic have already adapted their strategy, seeking to add piracy claims after discovering the shadow libraries it downloaded included books containing song lyrics.

While AI companies already defending suits weigh changes to their strategies, the settlement’s immediate impact may be a surge in new litigation. Cohen predicted a “rush of additional copyright lawsuits filed against Anthropic and others” now that the payment details are public.

The case is Bartz et al v. Anthropic PBC, N.D. Cal., 24-cv-05417, 9/5/25.