BRENNEN

Industries / Media, arts and recreation / Media and entertainment

Books and publishing

Book publishing is choosing, editing, making and selling books, in print, as ebooks and as audiobooks.

7

Milestones achieved, from 868 to 2026.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

In depth

The same ground is charted in more depth, as an industry of its own.

The road

AchievedWatch forOpen

Achieved

  1. 868

    The oldest dated printed book

    A copy of the Diamond Sutra, a Buddhist text, was printed in China from carved wooden blocks as a scroll, with a note saying Wang Jie had it made in what is 11 May 868 in the Western calendar. It was found in a cave at Mogao, near Dunhuang, around 1900, sold to the archaeologist Aurel Stein in 1907, and is now in the British Library. The Linda Hall Library calls it the oldest complete dated printed book known.

    lindahall.orgsmithsonianmag.com

    Credit

    Wang Jie, who paid for it; the unnamed block carvers

  2. c. 1455

    Gutenberg prints the Bible

    Johann Gutenberg and his partner Johann Fust published a Latin Bible in Mainz, Germany, in the mid-1450s, between 1454 and 1456 by the Harry Ransom Center's dating, printed with individual pieces of metal type. They printed more than 150 copies, and 20 complete copies survive. In the Ransom Center's words, the finished Bible "demonstrated the viability" of a press that used individual pieces of metal type.

    hrc.utexas.edu

    Credit

    Johann Gutenberg; Johann Fust

  3. 1710

    The first copyright law

    Britain's Statute of Anne gave the author of a new book, or whoever bought the rights, the sole right to print it for 14 years, plus 14 more if the author was still alive; books already in print got one 21-year term. The law scholar Ronan Deazley calls it "the world's first copyright statute," passed on 5 April 1710.

    LessonProtection was given a time limit, after which anyone could print the book, which is why a public domain exists.

    copyrighthistory.org

    Credit

    Parliament of Great Britain

  4. 1935

    Paperbacks for sixpence

    Allen Lane's new Penguin imprint put out ten paperbacks of well-regarded books, including Ernest Hemingway's A Farewell to Arms and Agatha Christie's The Mysterious Affair at Styles, on 30 July 1935 at sixpence each. The Design Museum notes that was the price of a packet of cigarettes, and that Lane insisted on plain covers colored by kind of book: orange for fiction, green for crime and blue for biography.

    guinnessworldrecords.comdesignmuseum.org

    Credit

    Allen Lane; The Bodley Head, which published the first ten

  5. 2007

    Books delivered over the air

    Amazon launched the Kindle e-reader at $399 with more than 90,000 books in its store, delivered in under a minute over a mobile data network with no phone contract and no computer needed.

    press.aboutamazon.com

    Credit

    Amazon

  6. 2015

    Scanning books ruled fair use

    Google scanned more than 20 million library books without permission to build a search index that shows short snippets. The US Court of Appeals for the Second Circuit ruled in Authors Guild v. Google that this was fair use, because search provides information about books without substituting for them.

    copyright.gov

    Credit

    Google; US Court of Appeals for the Second Circuit

  7. 2026

    Pirated training data paid for

    In Bartz v. Anthropic, Judge William Alsup ruled that training an AI model on lawfully bought books was fair use, but that keeping a library of books downloaded from pirate sites was not. Anthropic settled for $1.5 billion, about $3,000 for each of about 482,000 books, and Judge Araceli Martínez-Olguín gave final approval on 20 July 2026. Reports call it the largest known copyright recovery in US history.

    LessonHow a company gets its books mattered as much as how it used them; and because the case settled, no appeals court will review the fair-use ruling.

    techstrong.aijurist.org

    Credit

    the authors' class; Anthropic; US District Court for the Northern District of California

Watch for

  1. AI-read audiobooks for publishers

    Audible said it is working with selected publishers to turn print and ebooks into audiobooks read by AI voices, with more than 100 voices in English, French, Spanish and Italian, and a beta of AI translation from English into Spanish, French, Italian and German. TechCrunch found more than 50,000 titles on Audible already labeled "Virtual Voice," made with an earlier Amazon tool.

    Achieved whenAn outside party counts publisher titles narrated this way, or a listening test by someone other than Audible compares them with human narration.

    techcrunch.com

    Claimed by

    Audible, May 2025

Open

  1. 01

    Proofread old books by machine

    A pipeline that turns page scans of a public-domain book into a finished Project Gutenberg ebook. A model reads each page, a second pass compares the text with the page image, formatting such as italics and footnotes is marked, and a person checks only the pages the pipeline flags.

    The brief
    Credit · unclaimed
  2. 02

    Find the unrenewed books

    A checker for US books published before 1964 and still under the 95-year term. Such books lost their copyright if it was not renewed. For each scanned book in HathiTrust, a large library of digitized books, the checker searches the federal registration and renewal records, reads the scanned title and copyright pages for a notice, signs of foreign publication or material from other sources such as illustrations, and proposes "public domain" or "in copyright" with its evidence, for a human reviewer to accept or reject.

    The brief
    Credit · unclaimed

Open 01 · the brief

Proofread old books by machine

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A pipeline that turns page scans of a public-domain book into a finished Project Gutenberg ebook. A model reads each page, a second pass compares the text with the page image, formatting such as italics and footnotes is marked, and a person checks only the pages the pipeline flags.

Why it is still open

Distributed Proofreaders, the volunteer site that prepares many Project Gutenberg books, still has people proofread each book one page at a time through three levels of proofreading (P1 to P3), followed by formatting rounds. A 2025 study of open-weight AI models for correcting scanned text found they showed promise in reducing character errors in English but did not reach practically useful performance in Finnish. This research found no project that has produced finished Project Gutenberg books this way.

blog.pgdp.netarxiv.org

Built with what exists

Page scans of US public-domain books from the Internet Archive or HathiTrust; a vision-language model that reads page images, such as Claude Fable 5.2 (see the Note "Three working devices"), or the open-source Tesseract OCR engine; Python to compare outputs and flag pages.

  1. First step

    Take one book that Distributed Proofreaders finished, run its page scans through a vision model, and count the differences from the published text by script.

  2. Done looks like

    On three books Distributed Proofreaders has already finished, the pipeline's text, before any human check, differs from the published Project Gutenberg text in fewer than 1 character in 10,000, ignoring formatting; and one new book made this way is accepted and posted by Project Gutenberg.

Open 02 · the brief

Find the unrenewed books

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A checker for US books published before 1964 and still under the 95-year term. Such books lost their copyright if it was not renewed. For each scanned book in HathiTrust, a large library of digitized books, the checker searches the federal registration and renewal records, reads the scanned title and copyright pages for a notice, signs of foreign publication or material from other sources such as illustrations, and proposes "public domain" or "in copyright" with its evidence, for a human reviewer to accept or reject.

Why it is still open

The University of California's HathiTrust help pages (2022) say HathiTrust's "simple bibliographic algorithm cannot determine whether a volume was renewed or not" and that a knowledgeable person has to review each book. HathiTrust's volunteer reviewers had done more than 800,000 reviews since 2009 and opened more than 218,000 books, and the pages say hundreds of thousands of likely public-domain books remain closed because each has to be reviewed by hand. The New York Public Library has parsed the renewal records so that they can be matched to registrations automatically, but does not issue determinations. In a January 2025 report, the AI research group EleutherAI describes matching those records with a large language model for its Common Pile dataset and counts about 500,000 books it believes are in the public domain; it works from the copyright office's records, not from each scanned book, and says it is unclear how many have digital copies.

help.hathitrust.universityofcalifornia.edugithub.comarxiv.org

Built with what exists

NYPL's free registration and renewal data on GitHub; HathiTrust's Rights Database, which records reviewers' decisions, including "not renewed"; an AI model that reads page images for the copyright notice; Python.

  1. First step

    Take 100 US books from 1950 whose status HathiTrust's reviewers have already set, match each against NYPL's renewal file by script, and compare the result with the reviewers' decisions.

  2. Done looks like

    On 1,000 books HathiTrust's reviewers have already decided, the checker's "public domain" calls agree with theirs at least 98% of the time, it answers "unsure" rather than guessing on hard cases, and its evidence for each book is published.