Levine v. Meta Platforms, Inc.

One of five suits filed the same day

Levine is part of a wave. ChatGPT Is Eating the World reported that the same two firms filed five near-identical book-author suits against Meta on 19 October 2026: Sheard, Older, Winspear, Reno and Levine. By its count they cover 292 authors, with 61 in Levine. The next day it reported that the firms told Bloomberg Law they plan to sue Meta on behalf of 1,000 plaintiffs in the coming months. It also reported that the firms had earlier represented authors who opted out of the Bartz v. Anthropic class settlement.

The complaint explains the choice not to sue as a class. The plaintiffs say the Copyright Act entitles each of them to statutory damages set by a jury, and that they want to keep control of their claims rather than be swept into class settlements that resolve claims “for pennies on the dollar” (¶¶ 14-16). It cites Feltner v. Columbia Pictures Television for the rule that the amount of statutory damages is a jury question (¶ 15).

Who is suing

The caption and ¶ 1 name 61 plaintiffs. We counted them from the list in ¶ 1. Paragraphs 17 to 75 then introduce each of the 59 individual authors in turn, one paragraph each. Two of those paragraphs also name a company the author owns or controls: Nittany Valley Productions, Inc. for Paul Levine (¶ 17) and Mike Murphey Books, LLC for Mike Murphey (¶ 53). That gives 59 authors and two companies. ChatGPT Is Eating the World describes the case as 61 book authors.

Ten of the 59 authors are pleaded as living outside the United States, in Australia, Brazil, Canada, Colombia, Greece, Italy, Japan, Singapore, Switzerland and the United Kingdom. The books range from novels to technical and academic titles. Three footnotes flag errors in Copyright Office registration titles (nn. 3-5).

Four counts, two copying theories

Willfulness is pleaded for both direct counts. For Count I the complaint goes further and says the conduct “meets the standard for criminal copyright infringement” (¶ 164). For Count II it points to timing: Meta is said to have started Project Vellum after being sued over Llama, knowing the books carried copyright notices and that a licensing market existed (¶ 174).

Project Vellum

The scanning theory is the newer of the two. The complaint says Meta acquired print books in bulk, catalogued them by ISBN, title, author, publisher and year, scanned them, and ran OCR to extract the text (¶ 146). It says Meta used that text to train Muse Spark, released in April 2026, and Muse Glimmer, released to the open-source community in August 2026 (¶¶ 9, 147). The name “Project Vellum” comes from a Meta filing in another case (¶ 145 n.73).

The plaintiffs argue the scans were not “a one-for-one change of format” for Meta’s own reading, but wholesale copies kept in multiple versions to build a commercial training corpus (¶ 172). That framing matters because in Bartz v. Anthropic Judge Alsup held that Anthropic’s scanning of print books it had bought, replacing each print copy with a digital one, was fair use. The complaint does not cite Bartz on this point. Whether any plaintiff’s book was actually scanned is pleaded on information and belief (¶ 153).

The evidence the complaint relies on

Most of the specific facts come from Meta’s own statements in other cases. The complaint cites:

• Interrogatory answers served in Chicken Soup for the Soul, LLC v. Meta, No. 3:26-cv-02333 (N.D. Cal.). As quoted, Meta said Books3 (internally “b3g”) was used to train Llama 1 through 4; the LibGen “Scitech” and “Fiction” sets were used for Llama 3 and 4; and a Z-Library set, obtained by torrenting Anna’s Archive, was used for Llama 4 (¶ 142). Meta said its LibGen downloads were “believed to have been substantially complete” (¶ 142).
• The same answers on Project Vellum. Meta is quoted as admitting that scanned books were used to train the Muse Spark models, that their copyright text was not filtered out, and that it “has no feasible way to reliably ascertain” how many copies its pipeline made (¶¶ 147, 150-151). An Exhibit D to those answers is said to list the Muse Spark training books by ISBN, title, author, publisher and year (¶ 146 n.74).
• Meta’s answers in Kadrey and in Cambronne Inc. v. Meta, No. 3:26-cv-03725 (N.D. Cal.). The complaint quotes Meta as admitting a script built “to remove repetitive text” that was “designed to remove lines that contain the word ‘copyright’” (¶ 139), and that it used BitTorrent for some datasets (¶ 130 n.64).
• Judge Chhabria’s 2025 opinion in Kadrey v. Meta, 788 F. Supp. 3d 1026, for findings that Meta considered spending up to $100 million on licenses, then torrented LibGen and Anna’s Archive, and that an engineer wrote a script to prevent seeding but “apparently not leeching” (¶¶ 93, 126, 128, 130 n.65). The complaint also cites the Kadrey plaintiffs’ summary-judgment motion for the claim that Meta had downloaded over 25 TB from Z-Library and 10 TB from LibGen by April 2024 (¶ 129).

The work-level allegations are thinner. For each author, the complaint says one or more of the books appear in named pirate libraries “according to publicly available metadata,” and that there is “a reasonable inference” Meta downloaded them (¶¶ 17-75, 83). It does not say any specific plaintiff’s file was found in a dataset Meta used.

Damages

The complaint states no dollar figure and no count of works in suit. Exhibit A, the list of works, is described as non-exhaustive (¶ 76) and is not included in the posted 53-page complaint. Statutory damages under 17 U.S.C. § 504(c) run from $750 to $30,000 per work, and up to $150,000 per work for willful infringement. No total can be calculated until the number of works is known. The prayer also seeks statutory or actual damages for CMI removal under § 1203(c), an injunction, and an accounting of profits.

Relation to Kadrey v. Meta

On 29 October 2026, ChatGPT Is Eating the World reported that Judge Chhabria had related Hobbs, Older and Sheard to Kadrey v. Meta, so they will be reassigned to him. Levine was not on that list. The same post said it expected Winspear, Reno and Levine to follow. The sources reviewed for this page do not show an order relating Levine.