Separating AI hype from reality.

Publisher Sues Anthropic and Meta Together After Opting Out of Class Action Settlement

November 21, 2026 · Lawsuits · Anthropic · Meta

Carlson LLC filed a federal copyright infringement lawsuit on November 19 against Anthropic PBC and Meta Platforms, Inc. in the U.S. District Court for the Northern District of California, alleging both companies used pirated and scanned books to train their AI models without permission. The complaint, case number 3:26-cv-14248, names the two frontier labs in a single filing after Carlson LLC opted out of the Bartz v. Anthropic class action settlement [4]. A U.S. district judge approved that settlement in July 2026 [1][5].

The Bartz settlement and what Carlson left behind

The Bartz case, filed in August 2024 by a group of authors, alleged Anthropic "pirated" roughly 500,000 books for Claude's training data, sourcing them from Library Genesis, the Pirate Library Mirror, and the Books3 dataset [1]. Internal documents revealed a parallel effort called Project Panama, in which Anthropic purchased millions of books and chopped off their spines for rapid scanning [1]. In September 2025, Anthropic agreed to a settlement fund of at least $1.5 billion — the largest copyright settlement in U.S. history — with an estimated $3,000 per work before fees, split 50/50 between author and publisher [1][5]. Judge William Alsup approved the settlement in July 2026, though he drew a distinction: sourcing from pirated books violated fair use, but "the purpose and character of using copyrighted works to train LLMs to generate new text was quintessentially transformative" [1]. Because the case settled, that reasoning sets no binding precedent.

Carlson LLC, the registered copyright owner of literary works by R. and K. Carlson — including R. Carlson's 1997 book Don't Sweat the Small Stuff... and It's All Small Stuff (R. Carlson died in 2006) — chose not to participate [4]. By opting out, it preserves the right to pursue its own claims individually, a path Kyle Roche of Freedman Normand Friedland says his firm has also chosen for clients because "it can be difficult to prove that different class members were all affected by AI companies' conduct in the same way" [3].

What the new complaint alleges

The Carlson complaint alleges both Anthropic and Meta harvested training materials from pirate libraries Books3 and Anna's Archive, and acquired books by scanning physical copies without permission [4]. The allegations remain unproven; neither defendant has responded on the merits. Meta has previously acknowledged in internal correspondence, unearthed during the separate Kadrey v. Meta Platforms litigation, that clearing rights for training data was "unreasonably expensive" and "incredibly slow" [1]. Meta develops the Llama series of open models; Anthropic develops the Claude family.

A growing docket and a government thumb on the scale

The Carlson suit joins roughly 135 active copyright lawsuits against AI companies in the United States [1]. In September, the Department of Justice filed a brief in The New York Times v. Microsoft and OpenAI arguing that "the United States has a strong interest in this court rejecting any argument that training LLMs on copyrighted texts violates copyright law" [1][2]. That filing — the first time the U.S. government has weighed in on an AI-training copyright case — drew a letter from 12 Democratic House members on November 19 demanding to know why DOJ intervened without consulting the Copyright Office or Patent and Trademark Office, raising "serious questions about potential bias" [2]. A separate group of Judiciary Committee Democrats criticized the filing in October as contrary to standard practice [2].

Individual claims vs. collective leverage

Opting out of a class settlement shifts cost and risk to the rights holder. Wizards of the Coast, a participant in the Bartz settlement, announced November 20 it would forgo its publisher share and pass all proceeds to D&D authors — "hundreds of thousands of dollars" [5]. That decision came after author pressure, not legal compulsion. Carlson LLC's choice to sue individually tests whether a single publisher can extract more favorable terms — or establish precedent — outside the class framework. The Bartz settlement showed AI companies "may still face major copyright liability for unlawfully obtained training materials, even when AI model training itself may qualify as fair use," note Justin J. Gunnell and Benjamin J. Shack Sackler [3]. Whether Carlson's dual-defendant strategy changes that calculus remains to be seen.

Sources

  1. Return to Sender, by Benjamin Charles Germain Lee (harpers.org)
  2. Democratic lawmakers probe DOJ filing in OpenAI copyright case (reuters.com)
  3. Beyond Borders & Control: The Extension of U.S. Copyright ... (law.com)
  4. Publisher Sues Anthropic and Meta Over AI Training (wisevoter.com)
  5. Wizards of the Coast shares AI copyright settlement money with D&D's authors - Wargamer (wargamer.com)