# Bloomsbury to Collect on Anthropic’s $1.5B Book-Piracy Settlement, Court Confirms

By Dana Docket · 2026-07-22 · Licensing & Legal · https://datacommenter.com/bloomsbury-to-collect-on-anthropics-1-5b-book-piracy-settlement-court-confirms/
About the author: Legal correspondent for the data economy: scraping suits, copyright-and-AI litigation, privacy enforcement, and data-broker rules.
Original reporting: [Research Information](https://www.researchinformation.info/news/bloomsbury-to-share-in-1-5bn-anthropic-ai-copyright-settlement/)
_AI-assisted commentary, editorially reviewed. Quoted excerpts belong to the original outlet._

*A federal judge's final approval of the largest copyright settlement in US history means Bloomsbury will be paid for 14,087 titles Anthropic pirated to train Claude, at roughly $3,000 a book before fees.*

For publishers watching the generative-AI copyright wars, the question was never really whether AI companies would face liability for how they built their training sets — it was what a checkbook remedy would actually look like once the litigation dust settled. Now there’s an answer, and it has a line item with Bloomsbury’s name on it.

US District Judge Araceli Martínez-Olguín granted final approval Monday to the $1.5 billion settlement in *Bartz et al. v. Anthropic*, according to [Courthouse News](https://courthousenews.com/anthropic-to-pay-1-5-billion-copyright-settlement-to-authors-publishers/), closing out what plaintiffs’ attorney Justin Nelson called “the largest known copyright recovery in history.” Bloomsbury Publishing confirmed it is among the beneficiaries and expects payments to begin in the second half of its current financial year, according to [Research Information](https://www.researchinformation.info/news/bloomsbury-to-share-in-1-5bn-anthropic-ai-copyright-settlement/). The court has identified 14,087 Bloomsbury titles within scope, with compensation expected around $3,000 per title before legal fees and other deductions, split equally between publisher and author, Research Information reported.

### How the piracy case landed

The suit traced back to authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson, who alleged Anthropic built its Claude training corpus in part from millions of books downloaded from pirate repositories Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi), according to Research Information. Anthropic denied wrongdoing but agreed to settle rather than proceed to a trial that had been set for December 2025, Courthouse News reported. The legal footing beneath the settlement is more mixed than the headline number suggests: former Judge William Alsup, who granted preliminary approval before retiring, had already ruled that training AI models on copyrighted books counts as fair use, but found that Anthropic’s method of acquiring the books — mass downloads from known pirate sites — was independently unlawful, as [ABC News](https://abcnews.com/Technology/wireStory/judge-approves-15b-anthropic-settlement-pirated-books-train-134949964) and [TechCrunch](https://techcrunch.com/2026/07/20/anthropics-landmark-1-5b-copyright-settlement-is-approved/) both reported. That split ruling — fair use on training, liability on sourcing — is the reason Anthropic is paying $1.5 billion while still claiming a legal win on the underlying AI question.

### The numbers, and where they diverge

Accounts of the settlement’s scale vary slightly across outlets. Courthouse News put the payout at an estimated $3,100 per work across roughly 482,000 works and 54 known class members plus over 500,000 potential class members; ABC News cited more than 482,000 books and a per-book figure near $3,000; Ars Technica cited 506,194 works and noted the $3,000 figure works out to “four times the minimum statutory damages,” per Martínez-Olguín’s own order. Bloomsbury’s disclosed per-title figure of roughly $3,000 sits closer to the ABC/Ars figures than the Courthouse News number — the dossier does not resolve the discrepancy, which appears to stem from rounding or timing differences in court filings. Participation was high by any measure: about 95% of the class received notice and roughly 91% of authors and publishers had filed claims, according to Ars Technica, with only 350 class members opting out and 54 filing objections or late opt-outs.

### Lawyers took a haircut, too

Martínez-Olguín didn’t rubber-stamp everything plaintiffs’ counsel asked for. Lawyers initially sought 20% of the fund in fees — about $300 million — before reducing that ask to 12.5%, or roughly $187 million; the judge cut fees further, to just under 7% of the fund, or about $101.56 million, Ars Technica and Courthouse News both reported. She also trimmed the three named plaintiffs’ service awards from a requested $50,000 each to $15,000, finding the higher figure “unreasonable” absent evidence of retaliation. As a check on the reduced fee award, the court will require a post-distribution accounting once payouts are finalized, with fees subject to further reduction if the work proves lighter than projected, per Ars Technica. Anthropic, for its part, is also on the hook for non-monetary terms: it must destroy the original pirated files within 30 days of final judgment and certify that LibGen and PiLiMi datasets were not used in training any commercially released model, according to Courthouse News. Payment itself is staggered — $300 million paid shortly after preliminary approval, another $300 million due five days after Monday’s ruling, and $450 million installments on the first and second anniversaries of preliminary approval.

### What it doesn’t settle

For data companies and publishers alike, the deal’s limits matter as much as its size. Because Anthropic settled rather than appealed, Alsup’s fair-use finding never reached an appellate court and carries no binding precedent, TechCrunch noted — meaning every other AI copyright suit still has to be litigated on its own facts. Anthropic’s deputy general counsel, Aparna Sridhar, has framed the underlying ruling as establishing “that training AI on books is fair use under copyright law,” while stressing the company is “pleased that more than 91 percent of authors and publishers covered by the settlement have claimed their share of the payment.” That framing is a company position, not a settled legal question industrywide: TechCrunch reported that Hachette, Cengage, Elsevier, author Scott Turow and S.C.R.I.B.E. had just filed a separate class action against Google over Gemini training data, and similar suits continue against Meta, Midjourney and OpenAI. For rightsholders like Bloomsbury, the near-term story is a check in the mail; for the AI training-data economy broadly, the real test is whether the next judge to rule on scraped and pirated corpora reaches the same split verdict — fair use for the model, liability for the source.

> "The $1.5 billion settlement provides substantial benefits to the class in light of the novel claims asserted. Success at trial was not assured, and a loss would have left the class with no recourse."
> — [Research Information](https://www.researchinformation.info/news/bloomsbury-to-share-in-1-5bn-anthropic-ai-copyright-settlement/)

[Read the full story at Research Information →](https://www.researchinformation.info/news/bloomsbury-to-share-in-1-5bn-anthropic-ai-copyright-settlement/)

---

Cite this analysis: https://datacommenter.com/bloomsbury-to-collect-on-anthropics-1-5b-book-piracy-settlement-court-confirms/
Cite primary facts: https://www.researchinformation.info/news/bloomsbury-to-share-in-1-5bn-anthropic-ai-copyright-settlement/
Need the underlying datasets (alt data, market data, AI training data)? Source licensed vendors via Brickroad: https://brickroad.network
More machine-readable access: https://datacommenter.com/llms.txt

## Participate

- Comment on a passage: MCP `add_note` (include `source_url` when available).
- Suggest an editorially reviewed correction: MCP `suggest_edit`.
- Open factual questions: none.