Bartz v Anthropic PBC

'Bartz et al. v. Anthropic PBC (Case No. 3:24-cv-05417-WHA) is a landmark copyright class action lawsuit in which authors alleged that Anthropic trained its Claude large language models on millions of pirated books downloaded from shadow libraries. The case produced one of the most significant fair use rulings in AI copyright law and culminated in a $1.5 billion settlement — the largest U.S. copyright settlement to date — with a fairness hearing scheduled for May 14, 2026.

Field	Detail
Case Name	Bartz et al. v. Anthropic PBC
Court	U.S. District Court, Northern District of California (San Francisco)
Case Number	3:24-cv-05417-WHA
Judge	Hon. William Alsup
Filed	August 19, 2024
Plaintiffs	Andrea Bartz, Charles Graeber, Kirk Wallace Johnson (representing a class of authors)
Defendant	Anthropic PBC
Claims	Copyright infringement (piracy claims), class action
Status	$1.5B settlement preliminarily approved; fairness hearing May 14, 2026; 91.3% claim rate

Allegations

The plaintiffs alleged that Anthropic infringed their copyrights by downloading millions of pirated copies of copyrighted books from "shadow libraries" such as Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi) to compile a "central library" of "all the books in the world" for training its Claude large language models. They claimed this unauthorized reproduction for AI training displaced demand for their works, with no allegations of infringing outputs provided to users.

Fair Use Ruling (June 23, 2025)

On June 23, 2025, Judge Alsup issued a landmark summary judgment order holding that Anthropic's use of legally acquired books (including digitization of purchased print copies) for LLM training was "quintessentially" or "exceedingly" transformative and constituted fair use under Section 107 of the Copyright Act.

However, the court found that downloading and retaining pirated copies from LibGen/PiLiMi was infringing, as it displaced market demand and was not excused by later transformative use. This dual ruling — transformative fair use for lawfully acquired training data, but infringement for pirated copies — established a critical precedent in AI copyright law.

Class Certification (July 2025)

In July 2025, Judge Alsup certified a class action for piracy claims only (not AI training). The class was defined as all beneficial or legal copyright owners of reproduction rights in nearly 500,000 books from LibGen/PiLiMi datasets that had ISBN/ASIN numbers and were timely registered with the U.S. Copyright Office (within 5 years of publication or 3 months before Anthropic's download).

Settlement

In August 2025, Anthropic agreed to a $1.5 billion settlement, preliminarily approved after initial rejection of a proposal offering $3,000 per author. The settlement averted a December 1, 2025 trial on piracy liability, where statutory damages could have reached up to $150,000 per work — potentially exceeding $70 billion in total.

Key settlement terms:

• Settlement Amount: $1.5 billion total
• Scope: Releases Anthropic only for past acquisition, retention, and use of identified pirated works before August 25, 2025; excludes future training, AI outputs, or lawfully acquired materials, with no ongoing licensing requirement
• Claim Rate: 440,490 out of 482,460 eligible works (91.3%) were claimed by the March 30, 2026 deadline — far exceeding the typical 10% class action rate
• Fairness Hearing: Scheduled for May 14, 2026 at 2:00 p.m. (rescheduled from April 23, 2026, by Judge Martínez-Olguín's April 8 order)

Objections to the Settlement

Multiple objections were filed by the February 9, 2026 deadline, with several unsealed in April 2026 (Dkt. Nos. 544, 596, 598, 600, 601, 602, and subsequently Dkt. 630, 640, 641). Key objections include:

• Professor Bishop (Dkt. 630): Argues the settlement systematically favors publishers over authors — publishers could claim roughly 50% or more of the total settlement amount through royalty presumptions, despite the suit being author-driven. Also challenges the exclusion of over 2 million foreign and non-U.S.-registered works, which Bishop argues improperly narrows the class.
• Esquivel: Reports receiving inadequate notice, not learning of the settlement until approximately March 3, 2026 — nearly a month after the February 9 opt-out deadline had expired.
• Group registration undercounting (Dkt. 641): Objectors argue that multiple books registered under a single copyright registration are treated as one "claimable work," dramatically undercompensating prolific authors (e.g., 40+ books counted as a single work).
• Dangerous precedent (Dkt. 640): Argues the settlement allows AI companies to "buy their way out" of mass piracy at a discounted rate — cheaper than licensing — setting a harmful precedent for future AI copyright disputes.
• Class counsel conflicts: References Judge Alsup's December 2025 concerns about undisclosed fee-sharing arrangements between class counsel.
• Low per-work awards: Base awards of approximately $3,000 per work before deductions for attorneys' fees and costs, with a tiered system that pays more for "important" (nonfiction) books over fiction.

Objectors may participate in the May 14 fairness hearing via Zoom. Following the hearing, final approval and any appeals will determine when payments are distributed to class members.

Significance

The Bartz v. Anthropic ruling is one of the most consequential AI copyright decisions to date. Judge Alsup's dual framework — fair use for lawfully acquired training data but infringement for pirated copies — creates a clear template distinguishing between AI companies that license or legally obtain training data and those that rely on pirated sources. The $1.5 billion settlement amount underscores the significant financial risk AI companies face when using pirated materials for training.

The case also illustrates the high claim rates possible in copyright class actions involving AI: the 91.3% claim rate far exceeds typical class action rates, suggesting authors and rights holders are highly motivated to participate in AI-related copyright settlements.

Unsealed Objections and Pending Issues (April 2026)

Multiple objections to the $1.5 billion settlement were unsealed in April 2026, revealing significant concerns from class members:

• Exclusion of foreign works: Non-US-registered works are excluded from the class, potentially affecting over 2 million works.
• Distribution favoring publishers: The claims-made process and arbitration mechanism systematically benefit publishers over authors, with publishers potentially claiming roughly 50% or more of the total settlement amount — a stark departure for an author-focused lawsuit.
• Misleading class notice: The settlement notice uses "get nothing" framing that obscures the opt-out value (up to $150,000 per work in statutory damages versus approximately $3,000 per work in the settlement). One objector reported not receiving notice until March 3, 2026 — nearly a month after the February 9, 2026 opt-out deadline had expired.
• Inadequate class representation and counsel conflicts of interest: Objectors cite fee-sharing arrangements and allegations that Class Counsel omitted Judge Alsup's December 2025 concerns from updates to incoming Judge Martínez-Olguín.

The claim filing deadline passed on March 30, 2026. Additionally, Anthropic has filed a motion to sever and consolidate separate copyright lawsuits against it with the Bartz settlement, which would limit those authors' recovery to the guaranteed settlement amount rather than the up to $150,000 per work in statutory damages they are seeking.

As of April 2026, several objections remain sealed despite Judge Martínez-Olguín's unsealing order, including docket numbers 544, 596, 598, 600, 601, and 602.

See Also

• Anthropic $1.5B Settlement News Article
• Kadrey v Meta Platforms Inc — Parallel fair use ruling on AI training (Meta)
• Cases — Active AI litigation tracker
• Copyright Litigation — Category page