Apple Faces Consolidated Class Action Over AI Training on Authors' Books

A federal judge has consolidated three class action lawsuits accusing Apple of training its AI models on copyrighted books without permission, setting up one of the largest copyright fights in the technology sector. Judge Yvonne Gonzalez Rogers of the U.S. District Court for the Northern District of California approved the consolidation on November 14, 2025.

The lead case, Hendrix, et al. v. Apple Inc., was filed on September 5, 2025, under case number 3:25-cv-07558. Court records list two different docket prefixes — one pointing to the San Francisco Division, the other, 4:25-cv-07558-YGR, to the Oakland Division. The discrepancy between the records remains unresolved. What is not in dispute: the case sits before Judge Gonzalez Rogers.

What do the authors actually allege?

Plaintiffs Hendrix and Roberson, represented by Keller Rohrback LLP and Butterick Law P.C., say they and the proposed class hold registered copyrights in published works used to train Apple Intelligence models — specifically the Foundation Intelligence Models and OpenELM language models — without consent.

The complaint goes further than unauthorized use. It alleges Apple concealed the sources of its training datasets to avoid legal scrutiny, and that the company continues to hold a private AI training-data library that includes thousands of pirated books for future models, with no consent sought and no compensation offered.

The plaintiffs' core market argument, stated in their filing, is that Apple's models "compete with and dilute the market for those very works — works without which Apple Intelligence would have far less commercial value." That claim — that the books themselves create Apple's commercial value — is the theory the entire case will rise or fall on, and it is untested here.

The suit demands a jury trial, declaratory and injunctive relief, statutory and compensatory damages, restitution, disgorgement, and attorneys' fees. The class would cover copyright owners nationwide whose works were used for "Apple Intelligence Tasks" during the class period. Notably, the filings never define what "Apple Intelligence Tasks" means. That vagueness will matter when Apple's lawyers attack the class definition.

How big is this litigation wave?

The Apple cases sit inside a crowded docket. As of November 9, 2026, the tracker DataPacts counted 70 proceedings across 12 jurisdictions, spanning copyright, DMCA, contract, privacy, biometric and trademark claims tied to training data. Seventy cases in twelve courts is not a legal trickle; it is an industry-wide reckoning over the raw material of AI.

The precedents cut both ways, narrowly.

• In Bartz v. Anthropic (June 23, 2025), Judge William Alsup called generative AI training "exceedingly transformative" — but refused to excuse Anthropic's use of pirated library copies. The case then settled for a $1.5 billion fund covering 482,460 works.
• In Kadrey v. Meta (June 25, 2025), Judge Vince Chhabria granted Meta summary judgment because the thirteen author-plaintiffs made the wrong arguments and failed to build a record supporting the right one. He emphasized the ruling does not stand for the proposition that Meta's use of copyrighted materials to train its language models is lawful.
• On September 29, 2026, the Third Circuit held that ROSS Intelligence's non-generative legal research AI made "minimally transformative, at best" use of Westlaw headnotes and was not fair use — while stressing that ROSS "was not a generative AI," leaving the generative question open.

In plain terms: two district courts blessed generative training as fair use on their records, one appellate court rejected a non-generative use, and no court has definitively answered whether training generative models on copyrighted books is lawful. The Apple plaintiffs inherit that ambiguity.

What happens next?

The stipulated consolidation signals the court intends to manage the three cases jointly through discovery and pretrial motions. As of the tracker date of November 9, 2026, the docket shows no published schedule. The cases remain at the earliest possible stage: no motion to dismiss has been ruled on, no class certified, no discovery completed.

What remains unproven?

The allegation that Apple's training data includes thousands of pirated books is just that — an allegation. Discovery, if the case gets that far, will test it. Apple has every incentive to resolve a motion to dismiss first, and Kadrey showed how author-plaintiffs can lose on pleading failures alone.

The consolidated cases also leave open the actual function and scope of the Foundation Intelligence Models and OpenELM. The complaint names them as Apple Intelligence models and stops there.

The first substantive signal will be Apple's motion to dismiss — and the Bartz settlement, at $1.5 billion across 482,460 works, suggests that if the piracy allegations survive, Apple faces a price tag measured in the billions rather than the millions.