Background
Six authors — including journalist John Carreyrou (author of Bad Blood), Lisa Barretta, Michael Kochin, Philip Shishkin, Jane Adams, and Matthew Sacks — filed a copyright lawsuit in December 2025 against six major AI companies after opting out of a prior $1.5 billion class-action settlement. Their complaint alleged that Anthropic, Google, Apple, Nvidia, Perplexity AI, xAI, OpenAI, and Meta had copied the authors’ copyrighted books without permission from pirate libraries — including LibGen, Z-Library, and OceanofPDF — to train large language models. The authors sought $150,000 in statutory damages per copyrighted work per defendant.
The defendants moved to sever the claims, arguing that they should not be lumped together as co-defendants simply because they all allegedly trained LLMs on pirated data. Judge Pitts agreed to sever.
The Court’s Holding
Judge Pitts severed the suit and directed separate proceedings for each defendant under Federal Rule of Civil Procedure 21. The key finding: the complaint contained “no allegations of conspiracy” among the defendants, who were in fact “business rivals seeking to corner the LLM market.” Each defendant’s alleged copying of books to train its own AI was independent of the others — there was no coordinated scheme that would justify forcing them into a single lawsuit.
The case was split as follows: Anthropic remains in the Northern District of California as the sole defendant in the original action. Claims against Google, Apple, Nvidia, Perplexity AI, and xAI are severed into five separate lawsuits (each proceeding independently in N.D. Cal.). Claims against OpenAI are transferred to the In re: OpenAI Copyright Litigation multi-district litigation pending in the Southern District of New York.
Key Takeaways
- Plaintiffs cannot force multiple competing AI companies into a single lawsuit merely because each allegedly trained their models on the same pirated book repositories — independent acts of infringement by business rivals are not enough for joinder.
- Severance disperses the litigation cost but can also fragment the evidentiary record; authors pursuing parallel cases against six defendants will need to litigate each case on its own merits without sharing discovery across actions.
- Authors who opted out of the prior class settlement to seek individual statutory damages now face the burden of litigating six separate cases rather than a coordinated multi-defendant action — a significant strategic cost.
- The SDNY AI copyright MDL continues to centralize OpenAI-related claims; this ruling channels yet another major claim into that proceeding.
Why It Matters
The proliferation of AI copyright cases has led courts to grapple with the procedural mechanics of multi-defendant AI litigation. Judge Pitts’s severance ruling reflects a basic principle: each AI company built its LLM independently, and liability must be assessed defendant by defendant. This is a practical advantage for well-resourced AI companies, which can handle individual suits more comfortably than a consolidated action, and a potential disadvantage for individual authors who must sustain multiple separate litigations. The ruling also signals that class opt-outs pursuing maximum statutory damages face serious logistical hurdles in the multi-defendant AI copyright landscape.
Surfaced via Law360 IP.