Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson v. Anthropic PBC is a class-action copyright suit by authors against Anthropic, filed August 19, 2024 in the Northern District of California. The plaintiffs alleged that Anthropic used pirated books obtained from shadow libraries to train its Claude models. The case settled in 2025 for a reported ~$1.5 billion or more, the largest AI-training-copyright settlement reported to date; Judge Araceli Martínez-Olguín granted final approval on July 20, 2026.
Infobox
| Field | Value |
|---|---|
| Plaintiffs | Andrea Bartz, Charles Graeber, Kirk Wallace Johnson (on behalf of a class) |
| Defendant | Anthropic PBC |
| Court | N.D. Cal. |
| Filed | August 19, 2024 |
| Status | Settled (2025); final approval granted July 20, 2026 |
| Reported settlement | ~$1.5 billion or more (terms partially confidential) |
| Final approval hearing | May 14, 2026, San Francisco (Judge Araceli Martínez-Olguín) |
| Final approval | Granted July 20, 2026; class counsel fee cut from $187.5M requested to $101.6M |
Background
The plaintiffs are authors who alleged that Anthropic trained its Claude models on millions of copyrighted books obtained from pirated shadow-library sources, including Library Genesis (LibGen) and Z-Library. The suit is one of more than 105 active US AI copyright suits as of May 2026 (AI Copyright Litigation — Analysis).
Claims
The complaint alleged that Anthropic trained Claude on copyrighted books drawn from pirated shadow-library sources without authorization. The plaintiffs alleged willful infringement, for which statutory damages range from $750 to $150,000 per work. The proposed class covered all US authors whose books appear in the alleged training corpus.
Outcome
The case settled in 2025 for a reported ~$1.5 billion or more, with final terms partially confidential. Facing statutory-damages exposure across roughly 500,000 qualifying titles, Anthropic agreed to a settlement amounting to at least about $3,000 per title, funded in four installments through 2027; the court granted preliminary approval on September 25, 2025 (Source: authorsguild.org). Subsequent training-data suits, including Hachette et al. v. Meta (and Mark Zuckerberg), have been measured against the figure reported in Bartz as a reference point for AI-training-data copyright liability.
Settlement administration and final approval
The settlement's administration ran on the following schedule: an opt-out deadline of January 7, 2026 and a re-inclusion deadline of March 2, 2026 (Source: legalblogs.wolterskluwer.com), a claims deadline of March 30, 2026, and a final approval (fairness) hearing on May 14, 2026 in San Francisco (Source: anthropiccopyrightsettlement.com).
The fairness hearing was held May 14, 2026 before presiding Judge Araceli Martínez-Olguín. At the 75-minute hearing, seven objectors were each given two minutes; plaintiffs' lead attorney Justin Nelson reported that the settlement claims rate had risen from 91.3% to 92.77%, and there was no indication the roughly $3,000–$3,100-per-work payout would change. The judge's questions focused on attorneys' fees and the structure of the cost reserve rather than the merits of the objections, and she ordered Anthropic to file a supplemental brief by May 21 addressing why late opt-outs should not be honored (Source: publishersweekly.com; authorsalliance.org). Attorneys told the court that the opt-out and objection rates were "minuscule" (Source: publishingperspectives.com). As of June 2026, final approval remained pending, with observers expecting it to be granted (Source: clarkhill.com).
Judge Martínez-Olguín granted final approval on July 20, 2026. In the same order she reduced class counsel's requested $187.5 million fee to $101.6 million, applying a lodestar multiplier of 3.75 rather than the requested 6.92 (Source: chatgptiseatingtheworld.substack.com). Reuters described the approved settlement as the largest US copyright payout to date, roughly $3,000 per book across about 500,000 works (Source: reuters.com). Among identified beneficiaries, UK publisher Bloomsbury disclosed on July 22, 2026 that it will receive payouts covering 14,087 titles — roughly $3,000 each (Source: reuters.com).
Separately, on May 14, 2026, twenty-eight authors who had opted out of the settlement filed a new copyright suit against Anthropic requesting a jury trial, arguing that class-action treatment lets AI companies extinguish high-value copyright claims cheaply (Source: law.com).
The settlement is used as a reference point for AI-training-data copyright damages: the ~$1.5 billion or more figure has been cited as a practical-economic floor against which later settlements and judgments are compared. Bartz turned on the training-data question; the May 2026 Hachette et al. v. Meta (and Mark Zuckerberg) suit names Common Crawl and tests downstream-corpus-source liability. A separate question concerns AI-generated output rather than training inputs: in the April 2026 Claude Code source-code leak episode, Sigrid Jin used AI to rewrite leaked code in another language and Anthropic did not request a takedown, raising the question of whether AI-generated derivative work qualifies as transformative. Because Bartz settled rather than proceeding to verdict, it is also cited in considering whether settlement rather than verdict becomes the operative pattern across the 105+ active suits (AI Copyright Litigation — Analysis).
Relationships
- litigates: Anthropic.
- related: Hachette et al. v. Meta (and Mark Zuckerberg) (sibling training-data suit), NYT v. Microsoft, OpenAI et al. (earlier copyright suit anchor), Authors Guild V Openai (stub).
- related: AI Copyright Litigation — Analysis (umbrella).
- related: AI Content Licensing, AI and Tort Liability.
Sources
Provenance note: The complaint and settlement filings are queued for foundational ingest as primary texts; citations should upgrade to the primary documents once ingested.