Skip to content

Generated Source Mirror

This page is generated from the source Markdown file so it can be viewed inside the MkDocs site.

📝 Page Status: Draft — sourced but not yet reviewed

  • Status: draft
  • Sources: 3
  • Relationships: 4
  • Research debt items: 4

Bartz v. Anthropic PBC

Summary

Bartz v. Anthropic PBC, No. C 24-05417 (N.D. Cal.), is an authors’ class action over Anthropic’s copying of books to assemble a library and train Claude. On June 23, 2025, Judge William Alsup held that training on the books was exceedingly transformative fair use under 17 U.S.C. § 107, and that converting purchased print copies into digital library copies was also fair use, but that creating a permanent general-purpose library from pirated copies was not excused as fair use. On July 20, 2026, the court granted final approval of a $1.5 billion settlement of claims about past pirated acquisition and copying of works on a defined Works List.

Verified Facts

  • Named author plaintiffs include Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, with affiliated corporate entities Bartz Inc. and MJ + KJ Inc.1
  • Anthropic is described as an AI firm founded in January 2021 whose service Claude was first released publicly in March 2023.1
  • The fair-use order states that Anthropic downloaded Books3, LibGen, and PiLiMi collections containing pirated books, amounting to over seven million copies, and also purchased millions of print books that were stripped and scanned into a digital library.1
  • On summary judgment limited to fair use, the court held that using the books to train Claude and its precursors “was exceedingly transformative and was a fair use under Section 107.”1
  • The court held that digitizing purchased print copies for Anthropic’s central library was fair use because the digital copies replaced the purchased print copies without adding new copies, creating new works, or redistributing existing copies.1
  • The court held that Anthropic had no entitlement to use pirated copies for its central library and that creating a permanent, general-purpose library was not itself a fair use excusing the piracy.1
  • Authors did not allege in that motion practice that any LLM outputs infringed their works; the order addresses training copies and library copies.1
  • On July 20, 2026, the court granted final approval of a non-reversionary $1.5 billion settlement covering beneficial or legal copyright owners of books on a Works List drawn from LibGen or PiLiMi versions Anthropic downloaded, limited to works with ISBN or ASIN and timely U.S. copyright registration.2
  • The final-approval order states that the Works List contained 482,460 works, that about $3,000 per work was the estimated payment (four times the $750 ordinary statutory minimum), and that as of April 16, 2026, 440,490 works (91.3 percent) had been claimed.2
  • The court described the release as covering past AI inputs (pirating and copying) up to the point of any AI output, and as not releasing claims about past AI outputs or claims about future conduct on or after August 25, 2025.2
  • The settlement provides for destruction of class members’ pirated works; the order quotes Anthropic’s representation that neither the LibGen nor PiLiMi datasets were in the training corpus of any commercially released large language model.2
  • The court awarded class counsel $101,561,111 in fees (nearly 6.8 percent of the fund plus interest), expense reimbursement, and $15,000 service awards to each of three class representatives.2

Historical Context

The case was among the first to split AI-training copying from the method of acquiring the training corpus. Judge Alsup’s order is frequently contrasted with Kadrey v. Meta Platforms, Inc., decided two days later, which granted fair use on a different market-harm record and criticized Bartz’s treatment of market dilution.1

The later settlement is a class compromise of the remaining piracy-library theory, not a stipulated holding that training is always fair use or always infringement.2

Jurisdiction: U.S. District Court, Northern District of California. Authority level: district-court summary-judgment order plus class-settlement approval. Not circuit precedent.

The fair-use holding applies 17 U.S.C. § 107 factor by factor to three uses: (1) training copies, (2) purchased-and-scanned library copies, and (3) pirated library copies.1 Training was treated as transformative because the models learned to generate new text rather than to provide the books to the public as books.1 Pirated acquisition for a retained general-purpose library was treated as a substitute for paid copies and therefore not transformative in the same way.1 Buying a copy later does not, the court noted, absolve liability for earlier theft, though it may affect statutory damages.1

The settlement’s narrow release is legally significant: output claims and post-August 25, 2025 conduct remain outside the class deal, and works not on the Works List are unaffected.2

Relationships

  • CASE-BARTZ-V-ANTHROPIC cites SRC-BARTZ-V-ANTHROPIC-FAIR-USE.
  • CASE-BARTZ-V-ANTHROPIC related_to STAT-USC-107.
  • CASE-BARTZ-V-ANTHROPIC related_to TOPIC-AI-COPYRIGHT.
  • CASE-BARTZ-V-ANTHROPIC related_to CASE-KADREY-V-META.

Sources

  1. SRC-BARTZ-V-ANTHROPIC-FAIR-USE: Order on Fair Use (N.D. Cal. June 23, 2025).
  2. SRC-BARTZ-V-ANTHROPIC-FINAL-APPROVAL: Final approval order (N.D. Cal. July 20, 2026).

Additional sources (not yet cited in footnotes):

Research Debt

  • Add the class-certification order and settlement agreement as dedicated sources.
  • Confirm Ninth Circuit Rule 23(f) petition disposition.
  • Document opt-out litigation, if any, with docket citations.
  • Obtain an official PACER/RECAP copy of Dkt. 680 in addition to the DocumentCloud file.

Document metadata

  • Decision date: 2025-06-23
  • Last verified: 2026-08-22

  1. SRC-BARTZ-V-ANTHROPIC-FAIR-USE — Order on Fair Use (June 23, 2025). Primary district-court authority. ↩↩↩↩↩↩↩↩↩↩↩↩

  2. SRC-BARTZ-V-ANTHROPIC-FINAL-APPROVAL — Final approval order (July 20, 2026). Primary court order on settlement terms. ↩↩↩↩↩↩↩