PayAuthors white paper: dated pirate-library catalogs as evidence for which books AI companies acquired

What it is

An October 2026 PayAuthors white paper (text revised Oct. 14, 2026) on how dated pirate-library catalogs and U.S. copyright registrations can support research into which books AI companies acquired for AI development.

The data

• 22.4M catalog records across overlapping collections (not unique books)
• 31 dated catalogs and lists: 25 LibGen snapshots (Nov. 2015–Jun. 2025), 4 Z-Library catalogs (incl. two Pirate Library Mirror indexes from 2022), 1 Books3 list (196,608 file names; ISBN list cross-checked against file names), 1 Internet Archive lending-library catalog
• 8.03M U.S. literary-work registrations covered (1978–2025), 1.52M distinct registrations with at least one candidate catalog match

Key methodological points

A catalog can show that a library listed a particular edition by a particular date, and file identifiers or checksums can help compare a listing with files a company disclosed. It does not establish acquisition: that requires evidence such as a manifest, logs, testimony, or a stipulation. The paper grades evidence (court finding/order vs. company's own account vs. a party's allegation) separately from PayAuthors' own analysis, and notes that its exclusion screen for the Bartz settlement is an estimate, not the administrator's determination.

Case facts the paper draws on

• Bartz v. Anthropic: court-described downloads of millions of books from LibGen and the Pirate Library Mirror; the final Works List covered 482,460 works under the $1.5B settlement
• Meta has acknowledged acquiring LibGen and Anna's Archive data in its own filings
• A discovery order in the OpenAI litigation described an employee's 2018 LibGen download as undisputed
• Books3 (approx. 196,640 books, released 2020) was included in The Pile training dataset

This is a research guide, not a court document; it reports evidence and methods rather than asserting infringement.