PayAuthorscom

Preliminary (sample). Matches have passed the automatic checks and the software review; a person has not checked this sample’s matches. In a paid report, a person at PayAuthors checks the registration matches that affect the counts, and the reviewed version is added at the same link beside the preliminary one.

Sample report. This is what a Registration Details report looks like, built live for Andrea Bartz, a named plaintiff in Bartz v. Anthropic. Every paid report has this layout: a web page, a PDF and a CSV. Check your own name.

Author Presence Report

Andrea Bartz

Report SAMPLE
Prepared
This copy generated
Prepared by PayAuthors.com
Edition Registration Details (with the Copyright Office records)
Match name parts close together
Dataset payauthors-2026-10-01-m64, built Oct. 5, 2026
Analysis as of October 31, 2026
Sample rebuilt on each visit
Before you rely on this report (five checks)
  1. Untick any title that is not yours below; the PDF, CSV and counts follow your choice.
  2. Open each registration link and compare the title, author or claimant and edition with the Copyright Office record. “Record matched: other edition” means another edition of the same work.
  3. Treat lines listed under the surname only (or the surname and initials), lines found only by an ISBN that Open Library lists under the name searched, and matches marked “check this match”, as unconfirmed until you confirm them.
  4. Read “dates overlap or download date unknown” items as questions to investigate; they are never counted.
  5. The copyright claimant listed on a U.S. Copyright Office registration may not be the person who owns the work now. To know who owns it, review any written agreements with co-authors (if any) or a publisher, and consult a lawyer, who can also advise on transfers, timeliness and any settlement release.

Where your books were found

6Books
15Title lines
51Catalog records
5Registrations matched
4Books registered before a documented download

In short: 6 books under this name are in these datasets, plus 4 editions found only in another language, most likely translations of these books, marked “likely translation” and not counted as books. Together they appear as 15 title lines and 51 catalog records (copies, formats and editions). 5 of the books match a U.S. copyright registration (5 registrations in all), and 4 were registered before at least one documented download, the registration-date screen under 17 U.S.C. § 412. Click a title for its catalog records. Registration check: all 32 catalog-record matches to a registration confirmed against the Copyright Office record (26 same edition, 6 another edition of the same work; checked on title, author or claimant, edition and dates; one registration can match several catalog records).

Check these first

  • 1 catalog copy is a translation (another language, with a foreign title or ISBN), matched to the registration of the original: listed, not counted as the book’s presence in any dataset, because a translation is not the registered text; ask counsel whether it supports a claim: We were never here, Z-Library (after PiLiMi) record, Italian, 2024 (TX0009000657)
  • 4 catalog copies the library tags in another language carry the English title and an English ISBN (or none): counted as the registered book with a wrong language tag (confirm): The other side of the road, Z-Library (after PiLiMi) record, Portuguese, 2024 (TX0009377142); The spare room, Z-Library (after PiLiMi) record, Portuguese, 2023 (TX0009301448); The herd, Z-Library (after PiLiMi) record, Portuguese, 2020 (TX0008912519); The lost night, Z-Library (after PiLiMi) record, Portuguese, 2019 (TX0008858704)
LibGen7 titleslisted since Nov. 2015 (earliest snapshot showing one)
Z-Library14 titlesPiLiMi index (2022) and later catalogs to Jan. 2025
Books31 titleon the Books3 file list (2020)
Internet ArchiveNot found yetlending library (scanned print books)

File-level records

Each LibGen and Z-Library line names the exact file the library offered: its own ID, the MD5 fingerprint of the file’s exact bytes (what a lawyer matches against a company’s files in discovery), and the torrent and pieces that carried it.

  • 50files named by the library’s own ID, 35 with an MD5 fingerprint
  • 9found in a torrent’s own file list, which names the torrent; all with the piece layout; 10 PiLiMi release 2 files named by the PiLiMi index (whole .tar files) (no torrent is named for 2 files in ID blocks whose torrent we hold no copy of)
  • 1lies wholly inside one piece; 7 at least half

BitTorrent moves a collection in fixed-size blocks called pieces. When a book’s file fits inside one piece, anyone who received that piece received the whole book; whether one person sent all of it has to come from that person’s own records. Each file’s ID, MD5, torrent and piece layout is listed on the evidence page. What these mean.

TitleLibGenZ-LibraryBooks3Internet Archive (nonprofit library, not a pirate site)U.S. registrationPasses the date screen (§ 412)
The Lost NightWork ID W-W8CSN✓ since Feb. 2020✓ added April 2020✓—TX0008858704 Nov. 7, 2019 record PDF Record matchedAnthropic, Meta, DeepSeek, Bloomberg, Together AI
alleged: xAI (Grok), Perplexity, NVIDIA, Mark Zuckerberg (individual; dismissed from Sullivan without prejudice), Guillaume Lample (individual; claims dismissed without prejudice), Joelle Pineau (individual; claims dismissed without prejudice), Databricks / MosaicML, Salesforce, Apple, Cerebras, Adobe, Microsoft

dates overlap or unknown: Google (Gemini), OpenAI

Bartz Works List: listed, not on the opt-out list (released as to Anthropic)

Works List copy (TX0008858704): LibGen fiction ID 2157462, torrent f_2157000, pieces 238–241
We Were Never HereWork ID W-V000S · 2 lines, one work✓ since March 2022✓ added Aug. 2021——TX0009000657 Aug. 10, 2021 record PDF Record matchedMeta, DeepSeek
alleged: xAI (Grok), Google (Gemini), Perplexity, NVIDIA, Apple, Mark Zuckerberg (individual; dismissed from Sullivan without prejudice), Guillaume Lample (individual; claims dismissed without prejudice), Joelle Pineau (individual; claims dismissed without prejudice)

dates overlap or unknown: Anthropic

Bartz Works List: not found (TX0009000657)
The Spare RoomWork ID W-32S38✓ since Jan. 2024✓ added June 2023——TX0009301448 June 27, 2023 record PDF Record matchedMeta
alleged: NVIDIA

dates overlap or unknown: xAI (Grok), Google (Gemini), Perplexity, Apple, DeepSeek

Bartz Works List: not found (TX0009301448)
The HerdWork ID W-Q9WP4✓ since Feb. 2021✓ added Oct. 2019 before publication (March 2020)——TX0008912519 Nov. 12, 2020 record PDF Record matchedAnthropic, Meta, DeepSeek
alleged: xAI (Grok), Perplexity, NVIDIA, Mark Zuckerberg (individual; dismissed from Sullivan without prejudice), Guillaume Lample (individual; claims dismissed without prejudice), Joelle Pineau (individual; claims dismissed without prejudice)

dates overlap or unknown: Google (Gemini), Apple

Bartz Works List: listed, not on the opt-out list (released as to Anthropic)

Works List copy (TX0008912519): LibGen fiction ID 2212710, torrent f_2212000, pieces 782–783
We Were Never Here - Andrea BartzWork ID W-V000S · 2 lines, one work—✓ added June 2022——TX0009000657 Aug. 10, 2021 record PDF Record matchedMeta, DeepSeek
alleged: xAI (Grok), Google (Gemini), Perplexity, NVIDIA, Apple

Bartz Works List: not found (TX0009000657)
The Other Side of the Road (Never Tell collection)Work ID W-PNPF2 · 3 lines, one work✓ since June 2024✓ listed Jan. 2025——TX0009377142 March 29, 2024 record PDF Record matcheddates overlap or unknown: xAI (Grok), Google (Gemini), Perplexity, NVIDIA, Apple
Bartz Works List: not found (TX0009377142)
The other side of the roadWork ID W-PNPF2 · 3 lines, one work—✓ listed Jan. 2025——TX0009377142 March 29, 2024 record PDF Record matched: other editiondates overlap or unknown: xAI (Grok), Google (Gemini), Perplexity, NVIDIA, Apple
Bartz Works List: not found (TX0009377142)
The Other Side of the Road (Never Tell #3)Work ID W-PNPF2 · 3 lines, one work—✓ listed Jan. 2025——TX0009377142 March 29, 2024 record PDF Record matcheddates overlap or unknown: xAI (Grok), Google (Gemini), Perplexity, NVIDIA, Apple
Bartz Works List: not found (TX0009377142)
InnocentesWork ID W-DWRIJLikely translation · not counted as a book✓ since Jan. 2024✓ added June 2022————
Nunca Estivemos AquiWork ID W-PNI2RLikely translation · not counted as a book—✓ added May 2023————
Nigdy nas tu nie byłoWork ID W-VKAORLikely translation · not counted as a book—✓ listed Jan. 2025————
Flashback – Was hast du damals getan?: Thriller (German Edition)Work ID W-2357M · 3 lines, one work (same main title: confirm)Likely translation · not counted as a book—✓ added Aug. 2021————
Stuff Hipsters HateWork ID W-4LSFJ✓ since Nov. 2015—————
FlashbackWork ID W-2357M · 3 lines, one work (same main title: confirm)Likely translation · not counted as a book—✓ added Oct. 2021————
Flashback – Was hast du damals getan?Work ID W-2357M · 3 lines, one work (same main title: confirm)Likely translation · not counted as a book—✓ added July 2021————

Work ID (W-…) is PayAuthors’ own number for each book. Copies, formats and editions of one book share a Work ID and are counted once, because statutory damages are one award per work. A translation shares it when a registration or title ties it to the book; one found only in another language with no registration is marked “likely translation” and not counted as a book. The same ID is in the PDF and in the CSV’s work_id column.

Registration-date screen (§ 412) by possible defendant

Books in this report (all copies, editions and title lines of each) that were listed in a dataset by the time a company downloaded it, and carry a matched registration dated before the download, or registered within three months of a publication that came first (17 U.S.C. § 412). For the download date we use the later of the documented start of the download period and the library’s own record of when that copy was added (catalog metadata, not independent proof; for Books3, its October 2020 release); for an event known only by an outer bound (a paper, model release or complaint date), only the record date is available. A LibGen copy counts as in the dataset at a download only when a dated LibGen catalog snapshot taken on or before that download lists it; LibGen’s own “added” date is shown but is not enough on its own. A registration before that date passes, as does one made within three months of a publication that came first. Where the dates overlap or no download date is known, the comparison is listed separately and not counted. “Known from” says how the event is known. A date screen for investigation, not a finding that anyone copied a work or that statutory damages are available. Every event and its source is on the timeline of the AI copyright cases.

CompanyDataset · whenKnown fromBooksPasses the date screen
Anthropicreleased in the Bartz settlementBooks3early 2021 per the court record; on or around Feb. 25, 2021 per Anthropic's own answer (cutoff Feb. 24, 2021)court order1The Lost Night
Anthropicreleased in the Bartz settlementLibGenJune 2021 per the court record; June 8–19, 2021 per Anthropic's own answer (cutoff June 7, 2021)court order2The Herd, The Lost Night
Anthropicreleased in the Bartz settlementPiLiMiJuly 2022 per the court record; July 8 – Aug. 10, 2022 per Anthropic's own answer (cutoff July 7, 2022); part of PiLiMi: Anthropic took the PiLiMi copies of books it did not already have from LibGen, at least two million (Bartz Dkt. 244 at 2:25-3:5), so a book it already had from LibGen may not have been taken again from PiLiMicourt order2The Herd, The Lost Night
MetaLibGenOctober 2022; part of LibGen: Meta's verified interrogatory answers describe a partial torrent download of LibGen's non-fiction collection in October 2022 and a direct download in spring 2023, so whether a given copy was among the files taken is not showncourt order3The Herd, The Lost Night, We Were Never Here
MetaBooks3by Feb. 27, 2023 (LLaMA paper; outer bound)company's own paper or filing1The Lost Night
MetaAnna's Archive (LibGen, Z-Library and others)early 2024; part of Anna's Archive: which files Meta torrented is disputed; the court ordered Meta to produce its torrent servers' command histories (Kadrey Dkt. 831, Sept. 11, 2026)court order4The Herd, The Lost Night, The Spare Room, We Were Never Here
BloombergBooks3by Mar. 30, 2023 (BloombergGPT paper; outer bound)company's own paper or filing1The Lost Night
Together AIBooks3 (part, in the RedPajama dataset)by Apr. 17, 2023 (RedPajama release; outer bound); part of Books3: Together AI's answer says RedPajama's books section contained a subset of Books3, so whether a given copy was among them is not showncompany's own paper or filing1The Lost Night
DeepSeekAnna's Archive (e-books)by Mar. 8, 2024 (DeepSeek-VL paper; outer bound); not before DeepSeek’s lab was set up, April 2023; part of the archive: DeepSeek's paper describes 860,000 English and 180,000 Chinese e-books from Anna's Archive, so whether a given copy was among them is not showncompany's own paper or filing3The Herd, The Lost Night, We Were Never Here
Alleged or reported: leads for investigation
Mark Zuckerberg (individual; dismissed from Sullivan without prejudice)LibGen (Meta's torrenting)2022 as alleged; not before October 2022, when Meta first used LibGenallegation3The Herd, The Lost Night, We Were Never Here
Guillaume Lample (individual; claims dismissed without prejudice)LibGen (Meta's first copy, torrented)2022 as alleged; not before October 2022, when Meta first used LibGenallegation3The Herd, The Lost Night, We Were Never Here
Joelle Pineau (individual; claims dismissed without prejudice)LibGen (Meta's torrenting)2022 as alleged; not before October 2022, when Meta first used LibGenallegation3The Herd, The Lost Night, We Were Never Here
NVIDIABooks3 (via The Pile)2023 as allegedallegation1The Lost Night
Databricks / MosaicMLBooks3 (via RedPajama)by May 2023 (latest possible date: the MPT model release)allegation1The Lost Night
SalesforceBooks3 (via RedPajama / The Pile)by June 2023 (latest possible date: the xGen model release)allegation1The Lost Night
AppleBooks3 (via The Pile / RedPajama)by April 2024 (latest possible date: the OpenELM model release)allegation1The Lost Night
CerebrasBooks3 (via The Pile; SlimPajama)by Mar. 28, 2023 (latest possible date: the Cerebras-GPT release)allegation1The Lost Night
AdobeBooks3 (via SlimPajama / RedPajama)by Nov. 15, 2024 (latest possible date: the SlimLM paper)allegation1The Lost Night
Microsoftpirated books (about 200,000; dataset per complaint)date unknown (outer bound: complaint, June 2025)allegation1The Lost Night
xAI (Grok)LibGen, Z-Library, OceanofPDFdate unknown (outer bound: complaint, Dec. 2025); not before xAI was founded, March 2023allegation3The Herd, The Lost Night, We Were Never Here
Google (Gemini)LibGen, Z-Library, OceanofPDFdate unknown (outer bound: complaint, Dec. 2025)allegation1We Were Never Here†
PerplexityLibGen, Z-Library, OceanofPDFdate unknown (outer bound: complaint, Dec. 2025); not before Perplexity was founded, August 2022allegation3The Herd, The Lost Night, We Were Never Here
NVIDIAAnna's Archive (high-speed access)August 2023 as allegedallegation4The Herd, The Lost Night, The Spare Room, We Were Never Here
NVIDIALibGen, Z-Library (and Sci-Hub)date unknown (outer bound: amended complaint, Jan. 2026)allegation1We Were Never Here†
AppleAnna's Archivedate unknown (outer bound: amended complaint, Mar. 2026)allegation1We Were Never Here†

† registered within three months of publication, and published before the download.

No matched book passed this screen against: OpenAI, LibGen (2018); OpenAI, "Books1/Books2" (alleged LibGen; by May 2020); OpenAI, LibGen torrenting (alleged; 2019–20); OpenAI, Books3 (alleged; Feb. 2021); NVIDIA, Internet Archive lending books (alleged); Meta, Internet Archive books (alleged; by April 2024).

In the company’s own stated download window (registered, or added to the library, inside the days the company says it copied it: a possible claim, since whether the registration came before this copy was taken, or whether the copy was taken at all, depends on the day of the download, which the company’s records would show in discovery; shown, not counted): Anthropic, PiLiMi (July–Aug. 2022), window July 8 – Aug. 10, 2022 (1 book).

Dates overlap or download date known only as an outer bound (shown for investigation, not counted): OpenAI, "Books1/Books2" (alleged LibGen; by May 2020) (1 book); xAI, LibGen/Z-Library (alleged) (2 books); Google, LibGen/Z-Library (alleged) (4 books); Perplexity, LibGen/Z-Library (alleged) (2 books); DeepSeek, Anna's Archive (by March 2024) (1 book); NVIDIA, LibGen/Z-Library (alleged) (4 books); Apple, Anna's Archive (alleged) (4 books).

Registered after the same company’s earlier download of the book (not counted; § 412 runs from the company’s first download of a work): OpenAI, Books3 (alleged; Feb. 2021) (1 book).

Sources and cutoffs for each event
  • OpenAI, LibGen: 2018 per the court record; October–November 2018 as alleged (cutoff Sept. 30, 2018). Authors Guild v. OpenAI (S.D.N.Y.), discovery order Nov. 24, 2025 (Dkt. 782): undisputed that an employee downloaded LibGen books in 2018. Cutoff: class plaintiffs place the download in October and November 2018 (In re OpenAI, summary judgment memorandum at 5, Dkt. 1982, Sept. 17, 2026), so the cutoff is the day before, Sept. 30, 2018; the months are alleged, not found.
  • OpenAI, LibGen, as "Books1"/"Books2": by May 28, 2020 (GPT-3 paper; outer bound); not before OpenAI was founded, Dec. 2015. Class plaintiffs assert LibGen copies were renamed Books1 and Books2 (Authors Guild v. OpenAI, Dkt. 782); Books1/Books2 named in the GPT-3 paper (May 28, 2020).
  • OpenAI, LibGen, fiction and non-fiction (torrented): Sept. 2019 – Jan. 2020 as alleged. In re OpenAI Copyright Litigation (S.D.N.Y. 1:25-md-03143), class plaintiffs' Rule 56.1 statement ¶¶193–194 (Dkt. 1987, Sept. 17, 2026): about 35 TB, all LibGen fiction and non-fiction, Sept. 2019 – Jan. 2020.
  • OpenAI, Books3 (via The Pile): Feb. 1, 2021 as alleged (cutoff Jan. 31, 2021). In re OpenAI Copyright Litigation (S.D.N.Y. 1:25-md-03143), class plaintiffs' Rule 56.1 statement ¶¶404, 411–414 (Dkt. 1987, Sept. 17, 2026): an employee downloaded The Pile, 825 GB including Books3, into OpenAI storage on Feb. 1, 2021.
  • Anthropic, Books3: early 2021 per the court record; on or around Feb. 25, 2021 per Anthropic's own answer (cutoff Feb. 24, 2021). Bartz v. Anthropic, order of June 23, 2025 ("early 2021"). Cutoff: Anthropic's interrogatory answer No. 18 (Bartz Dkt. 554-6 at 4, served Apr. 22, 2025, filed Jan. 21, 2026; signed by counsel) says Ben Mann downloaded Books3 "on or around February 25, 2021", so the cutoff is the day before, Feb. 24, 2021; the day is the company's statement, not a finding.
  • Anthropic, LibGen: June 2021 per the court record; June 8–19, 2021 per Anthropic's own answer (cutoff June 7, 2021). Bartz v. Anthropic, order of June 23, 2025 (June 2021). Cutoff: Anthropic's interrogatory answer No. 18 (Bartz Dkt. 554-6 at 4, served Apr. 22, 2025, filed Jan. 21, 2026; signed by counsel) says the LibGen dataset was torrented "between June 8, 2021 and June 19, 2021", so the cutoff is the day before copying began, June 7, 2021; the days are the company's statement, not a finding.
  • Anthropic, PiLiMi: July 2022 per the court record; July 8 – Aug. 10, 2022 per Anthropic's own answer (cutoff July 7, 2022); part of PiLiMi: Anthropic took the PiLiMi copies of books it did not already have from LibGen, at least two million (Bartz Dkt. 244 at 2:25-3:5), so a book it already had from LibGen may not have been taken again from PiLiMi. Bartz v. Anthropic, order of June 23, 2025 (July 2022). Cutoff: Anthropic's interrogatory answer No. 18 (Bartz Dkt. 554-6 at 4, served Apr. 22, 2025, filed Jan. 21, 2026; signed by counsel) says the PiLiMi dataset was torrented by two employees "between July 8, 2022 and August 10, 2022", so the cutoff is the day before copying began, July 7, 2022; the days are the company's statement, not a finding. Aug. 10, 2022 is also the date the parties agreed to for the settlement. The same answer says Anthropic treats "books" as works with a 10- or 13-digit ISBN and does not know all the contents of Internet-sourced datasets such as components of The Pile and Common Crawl, so books in those datasets are outside its three-dataset answer.
  • Meta, LibGen: October 2022; part of LibGen: Meta's verified interrogatory answers describe a partial torrent download of LibGen's non-fiction collection in October 2022 and a direct download in spring 2023, so whether a given copy was among the files taken is not shown. Kadrey v. Meta, order of June 25, 2025 (Dkt. 598).
  • Meta, Books3: by Feb. 27, 2023 (LLaMA paper; outer bound). Kadrey v. Meta, order of June 25, 2025 (Llama training data included Books3); Touvron et al., LLaMA (Feb. 27, 2023).
  • Meta, Anna's Archive (LibGen, Z-Library and others): early 2024; part of Anna's Archive: which files Meta torrented is disputed; the court ordered Meta to produce its torrent servers' command histories (Kadrey Dkt. 831, Sept. 11, 2026). Kadrey v. Meta, order of June 25, 2025 (Dkt. 598 at 11–13): in early 2024 Meta downloaded Anna's Archive, a compilation including LibGen and Z-Library; undisputed that Meta torrented LibGen and Anna's Archive. Meta's Answer in Elsevier v. Meta (Dkt. 78 ¶97, July 13, 2026) admits obtaining Anna's Archive data in 2024.
  • Mark Zuckerberg (individual; dismissed from Sullivan without prejudice), LibGen (Meta's torrenting): 2022 as alleged; not before October 2022, when Meta first used LibGen. Elsevier v. Meta and Zuckerberg (S.D.N.Y. 1:26-cv-03689; transferred to N.D. Cal. Sept. 25, 2026), complaint ¶¶88–89 and contributory-infringement count against Zuckerberg; also named in Hobbs v. Meta (S.D.N.Y. 1:26-cv-04314) and Sullivan v. Meta (N.D. Cal. 3:26-cv-06793), from which he was dismissed without prejudice on Sept. 3, 2026 (Dkts. 60, 62). Cutoff: Meta first used LibGen in October 2022 (Kadrey v. Meta, order of June 25, 2025) and its answers date its LibGen download to in or about October 2022 (Sullivan v. Meta, Dkt. 85 ¶49), so a 2022 download cannot predate October; the cutoff is the day before, Sept. 30, 2022.
  • Guillaume Lample (individual; claims dismissed without prejudice), LibGen (Meta's first copy, torrented): 2022 as alleged; not before October 2022, when Meta first used LibGen. Hobbs v. Meta, Zuckerberg, Lample and Pineau (S.D.N.Y. 1:26-cv-04314), complaint ¶¶33, 66 (May 22, 2026): while at Meta, Lample torrented a copy of LibGen in 2022, including LibGen non-fiction; he is now a co-founder of Mistral AI. The plaintiffs dismissed their claims against Lample without prejudice (notice Sept. 9, approved Sept. 11, 2026), so they could be brought again; the amended complaint of Sept. 16, 2026 names Meta, Zuckerberg and unnamed defendants. Cutoff: Meta first used LibGen in October 2022 (Kadrey v. Meta, order of June 25, 2025) and its answers date its LibGen download to in or about October 2022 (Sullivan v. Meta, Dkt. 85 ¶49), so a 2022 download cannot predate October; the cutoff is the day before, Sept. 30, 2022.
  • Joelle Pineau (individual; claims dismissed without prejudice), LibGen (Meta's torrenting): 2022 as alleged; not before October 2022, when Meta first used LibGen. Hobbs v. Meta, Zuckerberg, Lample and Pineau (S.D.N.Y. 1:26-cv-04314), complaint ¶32 (May 22, 2026): named as a defendant as the Meta VP who led FAIR and oversaw Llama; no separate download date is alleged, so the cutoff follows Meta's 2022 LibGen download. The plaintiffs dismissed their claims against Pineau without prejudice (notice Sept. 9, approved Sept. 11, 2026), so they could be brought again; the amended complaint of Sept. 16, 2026 names Meta, Zuckerberg and unnamed defendants. Cutoff: Meta first used LibGen in October 2022 (Kadrey v. Meta, order of June 25, 2025) and its answers date its LibGen download to in or about October 2022 (Sullivan v. Meta, Dkt. 85 ¶49), so a 2022 download cannot predate October; the cutoff is the day before, Sept. 30, 2022.
  • Bloomberg, Books3: by Mar. 30, 2023 (BloombergGPT paper; outer bound). Wu et al., BloombergGPT (Mar. 30, 2023); Huckabee v. Bloomberg (S.D.N.Y.), motion to dismiss denied Nov. 24, 2025.
  • NVIDIA, Books3 (via The Pile): 2023 as alleged. Nazemian v. NVIDIA (N.D. Cal.), complaint; direct-infringement claim survived dismissal May 2026.
  • Databricks / MosaicML, Books3 (via RedPajama): by May 2023 (latest possible date: the MPT model release). O'Nan v. Databricks / In re Mosaic LLM Litigation (N.D. Cal. 3:24-cv-01451), complaint.
  • Salesforce, Books3 (via RedPajama / The Pile): by June 2023 (latest possible date: the xGen model release). Tanzer v. Salesforce, class complaint (Oct. 2025).
  • Apple, Books3 (via The Pile / RedPajama): by April 2024 (latest possible date: the OpenELM model release). Hendrix v. Apple (N.D. Cal., Sept. 2025); Martinez-Conde v. Apple (N.D. Cal., Oct. 9, 2025), complaints.
  • Together AI, Books3 (part, in the RedPajama dataset): by Apr. 17, 2023 (RedPajama release; outer bound); part of Books3: Together AI's answer says RedPajama's books section contained a subset of Books3, so whether a given copy was among them is not shown. James v. Together Computer, Inc. (N.D. Cal. 3:25-cv-09578), Together AI's answer to the first amended complaint (Dkt. 54, June 3, 2026) ¶¶22–23, 29, 31–33: it is one of the collaborators that created RedPajama; as released in April 2023 the dataset's books section contained a subset of Books3; its Apr. 17, 2023 blog post linked the download; Books3 was removed Apr. 12, 2024. The answer does not admit training Together AI's own models on the books section..
  • Cerebras, Books3 (via The Pile; SlimPajama): by Mar. 28, 2023 (latest possible date: the Cerebras-GPT release). James v. Cerebras Systems Inc. (N.D. Cal. 4:25-cv-09361), complaint ¶¶26–28, 30 (Oct. 30, 2025): Cerebras-GPT pre-trained on The Pile, including Books3. Cerebras's answer to the amended complaint (Dkt. 76, July 9, 2026) ¶¶28, 33, 39, 62 admits using portions of The Pile for Cerebras-GPT and building its SlimPajama dataset from portions of RedPajama; it does not admit Books3..
  • Adobe, Books3 (via SlimPajama / RedPajama): by Nov. 15, 2024 (latest possible date: the SlimLM paper). Lyon v. Adobe Inc. (N.D. Cal. 3:25-cv-10732), complaint (Dec. 2025): the SlimLM models were pre-trained on SlimPajama, built from RedPajama, whose books part is Books3. Adobe's answer to the consolidated complaint (Dkt. 36, May 22, 2026) ¶¶5, 27, 28 admits SlimPajama was used to pre-train its SlimLM research models and does not mention Books3; SlimLM paper, arXiv 2411.09944 (Nov. 15, 2024)..
  • Microsoft, pirated books (about 200,000; dataset per complaint): date unknown (outer bound: complaint, June 2025). Bird v. Microsoft (S.D.N.Y., June 2025), complaint (Megatron).
  • xAI (Grok), LibGen, Z-Library, OceanofPDF: date unknown (outer bound: complaint, Dec. 2025); not before xAI was founded, March 2023. Carreyrou et al. v. Anthropic et al. (N.D. Cal. 5:25-cv-10897), complaint ¶2 (Dec. 22, 2025); the amended complaint (Dkt. 119, Mar. 10, 2026) ¶39 adds a 'reasonable basis to believe' xAI downloaded Anna's Archive.
  • Google (Gemini), LibGen, Z-Library, OceanofPDF: date unknown (outer bound: complaint, Dec. 2025). Carreyrou et al. v. Anthropic et al. (N.D. Cal. 5:25-cv-10897), complaint ¶¶2, 8, 74 (Dec. 22, 2025): Gemini and Imagen alleged to rely on Z-Library and OceanofPDF data.
  • Perplexity, LibGen, Z-Library, OceanofPDF: date unknown (outer bound: complaint, Dec. 2025); not before Perplexity was founded, August 2022. Carreyrou et al. v. Anthropic et al. (N.D. Cal. 5:25-cv-10897), complaint ¶2 (Dec. 22, 2025); amended complaint ¶40 (Dkt. 119, Mar. 10, 2026) adds a 'reasonable basis to believe' Perplexity downloaded Anna's Archive.
  • DeepSeek, Anna's Archive (e-books): by Mar. 8, 2024 (DeepSeek-VL paper; outer bound); not before DeepSeek’s lab was set up, April 2023; part of the archive: DeepSeek's paper describes 860,000 English and 180,000 Chinese e-books from Anna's Archive, so whether a given copy was among them is not shown. Lu et al., DeepSeek-VL (arXiv 2403.05525, v1 Mar. 8, 2024), §2.1: 860K English and 180K Chinese e-books from Anna's Archive, cleaned for document OCR data.
  • NVIDIA, Anna's Archive (high-speed access): August 2023 as alleged. Nazemian v. NVIDIA (N.D. Cal. 4:24-cv-01454), first consolidated amended complaint ¶¶53–57 (ECF 235, Jan. 16, 2026): NVIDIA contacted Anna's Archive about high-speed access in August 2023 and management approved within a week. NVIDIA withdrew its motion to dismiss as to Anna's Archive (ECF 271, Mar. 6, 2026); its Answer admits downloading data for AI research from annas-archive.org/datasets and libgen.rs, without dates, and denies improperly copying the plaintiffs' works (ECF 321 ¶¶58, 77, June 9, 2026).
  • NVIDIA, LibGen, Z-Library (and Sci-Hub): date unknown (outer bound: amended complaint, Jan. 2026). Nazemian v. NVIDIA (N.D. Cal. 4:24-cv-01454), first consolidated amended complaint ¶58 (ECF 235, Jan. 16, 2026), on information and belief; Carreyrou amended complaint ¶41 (Dkt. 119, Mar. 10, 2026) alleges NVIDIA downloaded Anna's Archive.
  • NVIDIA, Internet Archive lending books (in Anna's Archive's offer): August 2023 as alleged. Nazemian v. NVIDIA, first consolidated amended complaint ¶57 (ECF 235, Jan. 16, 2026): Anna's Archive's offer, about 500 TB, included millions of Internet Archive lending books.
  • Meta, Internet Archive books (46 TB, as alleged): by mid-April 2024 as alleged. Hobbs v. Meta (S.D.N.Y. 1:26-cv-04314), complaint ¶¶73–74 (May 22, 2026): by mid-April 2024 Meta held 46 TB from the Internet Archive (5.8 million files), 25.7 TB from Z-Library and 10 TB from LibGen.
  • Apple, Anna's Archive: date unknown (outer bound: amended complaint, Mar. 2026). Carreyrou et al. v. Anthropic et al. (N.D. Cal. 5:25-cv-10897), amended complaint ¶38 (Dkt. 119, Mar. 10, 2026): a 'reasonable basis to believe' Apple downloaded Anna's Archive.

Books registered before each company’s download. Statutory damages are one award per work for all of one company’s infringements in a case (17 U.S.C. § 504(c)(1)), so the count that matters is books, company by company.

  • Meta4 booksregistered before a documented download
  • DeepSeek3 booksregistered before a documented download
  • Anthropic2 booksregistered before a documented downloadReleased in the Bartz settlement: Andrea Bartz is a named plaintiff, so her works on the Works List cannot be claimed again against Anthropic.
  • Bloomberg1 bookregistered before a documented download
  • Together AI1 bookregistered before a documented download
  • NVIDIA4 booksregistered before an alleged download
  • Guillaume Lample (individual; claims dismissed without prejudice)3 booksregistered before an alleged download
  • Joelle Pineau (individual; claims dismissed without prejudice)3 booksregistered before an alleged download
  • Mark Zuckerberg (individual; dismissed from Sullivan without prejudice)3 booksregistered before an alleged download
  • Perplexity3 booksregistered before an alleged download
  • xAI (Grok)3 booksregistered before an alleged download
  • Apple2 booksregistered before an alleged download
  • Adobe1 bookregistered before an alleged download
  • Cerebras1 bookregistered before an alleged download
  • Databricks / MosaicML1 bookregistered before an alleged download
  • Google (Gemini)1 bookregistered before an alleged download
  • Microsoft1 bookregistered before an alleged download
  • Salesforce1 bookregistered before an alleged download

Not counted: 1 book whose registration dates overlap a download period or where the download date is not known.

This sample shows counts, not dollar amounts. Whether any book qualifies, and for what, is for a court. Litigation support, not legal advice.

The full Method and limits and Data provenance (every source file with its SHA-256 checksum) are on the evidence page and in the PDF.

What this report shows, and what it does not

This report compares the name searched with the dated Library Genesis catalog snapshots listed under Data provenance, the PiLiMi index of Z-Library, later Z-Library catalog records (Anna’s Archive’s index of records added August 2022 to August 2023, and the MajinBook extract of the January 2025 catalog), the Internet Archive’s lending-library catalog, the list of Books3 file names published by Peter Schoppert, and U.S. Copyright Office registration data as published in January 2026, each identified by version and checksum in this report. A match means that a corresponding catalog record or file name appears in that source. It does not establish that any person or company downloaded, copied or used the listed file or dataset, or that any model was trained on a work. The Internet Archive is a nonprofit library, not a pirate site; a lending-library line means the Archive lists a scanned copy of the book for borrowing. A book counts as in Books3 only when a file in the Books3 file list carries it; an ISBN printed somewhere in the Books3 texts is not enough, because a book’s text prints other books’ ISBNs, and the report says when a list that goes by those ISBNs would be wrong. Copyright Office results are candidate matches to public registration records; each line states how the match was made and links to the Office’s record for verification. A match is not a legal determination of authorship, ownership or infringement, and the absence of a match does not show that a work is unregistered or absent from another dataset. LibGen catalog entries for the same edition are combined into one line with a count; no entry is dropped. Matching is automated and not perfect: expect some missed records and some matches that are not the author’s, and check any line you rely on. Source-file checksums identify the exact files analyzed; they do not, by themselves, establish a file’s origin, completeness or authenticity. This report is litigation support, not legal advice.

Plain-English explanation of what a match does and does not show: payauthors.com/what-a-match-shows. Sources, checksums and matching methods: payauthors.com/methodology.