daniel@home:~$

Shadow libraries: the library nobody manages to close

Picture the scene. You’re nineteen, studying far from home, and the professor assigns a chapter your faculty library doesn’t own. On the publisher’s site it costs €42. The paper you need for your term paper, €38. It’s not that you don’t want to pay: it’s that you can’t. And the person who wrote that paper isn’t getting paid for it either — their university wrote it, on a public grant, and published it there because there was no other option.

Someone solved this twenty years ago. They made a copy, catalogued it, and shelved it on a wall-free bookshelf anyone with a connection can walk into. That is a shadow library (also called pirate libraries or “black open access”): online repositories that hand out for free what normally sits behind a paywall.

A warning before going further: downloading copyrighted works without permission is illegal in most of the world, and this is not a guide to doing it. This is the story of a library that exists, of why it exists, and of why nobody — not the FBI, not Cloudflare, not a dozen court rulings — has managed to switch it off.

A tradition older than the internet

The phenomenon didn’t start with torrents. It starts in the Soviet Union, where printing was controlled and dissident material was copied by hand and passed from person to hand: samizdat. When the Wall came down the habit didn’t disappear; the paper did. In 1994 writer Maksim Moshkow put up lib.ru, a collection of digitised texts that became a reference for a whole Russian generation. In parallel, a group of volunteer librarians and scientists — the Kolkhoz collection — used borrowed institutional passwords to download papers and stored them on FTP servers and DVDs: some 50,000 documents.

From there come the first big ones: Gigapedia / Library.nu, around 2010 the largest shadow library in the world, shut down in 2012 after a lawsuit by seventeen publishers. And above all Library Genesis (LibGen, circa 2008), built by Russian scientists. LibGen introduced the idea that changed everything: instead of hiding, it published its catalogue, its data and its source code so anyone could mirror it. It absorbed what was left of Library.nu and became the shared infrastructure for half the internet.

Then came the rest: Sci-Hub (5 September 2011, Alexandra Elbakyan, from Kazakhstan, with a system that fetched a paper on request and served it), Z-Library (which began as a LibGen mirror and grew past it) and Anna’s Archive (November 2022), which is less a library than a search engine across all of the above, with the data packaged as torrents so it doesn’t depend on any single server.

The numbers, out loud

Project What it holds (per its own catalogue) What has been thrown at it
Library Genesis (2008) 2.4m non-fiction books, 80m articles $30m judgment for Pearson, McGraw Hill, Macmillan and Cengage (2024); most of its domains seized between 2024 and 2025
Sci-Hub (2011) 88.3m papers (2022) $15m default judgment for Elsevier (2017); uploads frozen since 2020
Z-Library (2009) 13.35m books, 84.8m articles (2023) 240+ domains seized by DOJ and FBI (2022); its two founders arrested in Argentina, facing up to 20 years
Anna’s Archive (2022) 71,996,952 books and 239,545,528 papers; roughly 1.1 petabytes in torrents $322m judgment (Spotify and the labels, April 2026) plus another $19.5m with a global domain-takedown order (May 2026)

To put the second column in perspective: in March 2025 Anna’s Archive averaged 650,000 downloads a day, roughly ten times what the New York Public Library distributes daily. And a 2018 study published in eLife estimated that Sci-Hub provided access to nearly all paywalled scholarly literature in existence.

Why are they still there? The uncomfortable part

Because there is real, massive, badly-served demand. It’s the so-called serials crisis: libraries have been cutting subscriptions for decades because prices rise far faster than inflation, while the work itself — publicly funded research, peer-reviewed for free, written by salaried academics — is done by other people. RELX, Elsevier’s parent company, closed 2025 with an adjusted operating margin of 34.8%. It’s a business of renting out keys.

LibGen’s own admins said it bluntly, in words recorded by researcher Jerzy Karaganis: “the target groups for LibGen are poors: Africa, India, Pakistan, Iran, Iraq, China, Russia and post-USSR, and on a separate note, people who do not belong to academia. If you are not at a university, you can’t access anything.” Elbakyan frames it as a right: Article 27 of the Universal Declaration speaks of freely participating in science and its benefits, hence her motto, “any law against knowledge is fundamentally unjust”. Z-Library put it far more simply: “not everyone has access to a library, and not every library in the world is well-stocked.”

And then there’s the text that became the anthem of all this, the Guerilla Open Access Manifesto Aaron Swartz wrote in 2008: “The world’s entire scientific and cultural heritage, published over centuries in books and journals, is increasingly being digitized and locked up by a handful of private corporations… Those with access to these resources — students, librarians, scientists — you have been given a privilege. You get to feed at this banquet of knowledge while the rest of the world is locked out. But you need not — indeed, morally, you cannot — keep this privilege for yourselves.”

Now the genuinely uncomfortable part, because it has one. Those shelves don’t hold only 2009 Elsevier papers: they hold novels by living people who earn fourteen euros a copy, manuals written by professionals, translations someone did for free and others someone did for money. That the middleman is a monster with 35% margins doesn’t make the writing free. Dave Hansen, of the Authors Alliance, put it precisely: shadow libraries are “a kind of symptom of how broken the system is, particularly when you’re looking at access to scientific articles”. A symptom, not the cure.

The crackdown, which looks like a hydra

The enforcement record is long and, oddly, instructive.

In November 2022 the United States seized more than 240 Z-Library domains and arrested its two operators in Argentina. Z-Library came back: first over Tor, then with browser extensions and individual access domains issued to each registered user — a strategy designed precisely so that seizing one domain accomplishes nothing.

LibGen was beheaded through civil litigation: $30m in 2024, mass domain seizures, and in Germany a nationwide block coordinated by the CUII — an alliance of rightsholders and ISPs — applied without a court order, after consulting the telecoms regulator. The Netherlands has had a “dynamic” blocking order against LibGen and Anna’s Archive since 2024: if the sites change domains, providers must update the block automatically. India banned Z-Library in 2022.

Anna’s Archive has had everything thrown at it: Telegram closed its channel in January 2025; OCLC sued it for copying the entire WorldCat catalogue and in January 2026 won a ruling ordering that data deleted; after the mass extraction of Spotify’s music (about 300 terabytes, 86 million tracks, 37% of the catalogue but 99.6% of all listens) came the April 2026 judgment of $322 million, and with it Cloudflare disabling its nameservers, which pushed the site into the arms of a Russian anti-DDoS provider. In May 2026, thirteen publishers and the Association of American Publishers won another $19.5m plus a global domain takedown order.

And it still works. It still works because the copy no longer lives on one machine: it lives in 1.1 petabytes of torrents replicated across tens of thousands of drives, and because the model is contagious enough that clones have appeared which copy even its source code. The lesson is awkward for both the pursuers and the defenders: blocking works on people, not on data.

The twist nobody saw: the shadow fed the AIs

This is what changed the board between 2025 and 2026.

Documents unsealed in novelists Richard Kadrey and Sarah Silverman’s lawsuit against Meta revealed the company had torrented 81.7 terabytes of books, some 35.7 from Z-Library and LibGen via Anna’s Archive, to train Llama. The Atlantic built a search tool over the leaked LibGen database so any author could check whether they were in it. Anthropic did something similar at larger scale: according to the evidence, it used LibGen to obtain at least five million books, and in September 2025 agreed to pay $1.5 billion to authors, roughly $3,000 per work — the largest copyright settlement in US history. Nvidia, for its part, faces a 2024 lawsuit over the “The Pile” corpus, in which evidence emerged in January 2026 that it had contacted Anna’s Archive directly to obtain high-speed access to its data.

As early as 2025 Anna’s Archive admitted giving that access over SFTP to some thirty companies, mainly Chinese, in exchange for money or data. In other words: the network built so a student in Pakistan could read a textbook ended up as the raw-material supplier of the richest companies on the planet. And the only parties who have actually paid for it are those companies, and not out of conviction: because it was cheaper than the lawsuit. If that is a victory for access to knowledge, somebody will have to explain it to me slowly.

What you can do without crossing the line

That the paywall is absurd doesn’t mean the only door left is the back one. There are legal doors, and almost nobody uses them:

  • Unpaywall and Open Access Button: extensions that, when you hit a paywalled paper, look for a legal free version self-archived by the author. They work far more often than people think.
  • Open repositories: arXiv (physics, maths, AI), PubMed Central (biomedicine), the DOAJ (open-access journals), SciELO and Redalyc (excellent Ibero-American coverage, in Spanish).
  • Public domain and lending: Project Gutenberg, the Internet Archive and its Open Library, the Biblioteca Virtual Miguel de Cervantes, and if you have a Spanish public library card, eBiblio: free ebook and audiobook lending through an app, via your own town.
  • Beware the fakes: the biggest real risk in that world isn’t legal, it’s security. In 2024 the personal data of nearly ten million users collected by a fake Z-Library clone leaked. A site like this is a perfect target for phishing and for tampered PDFs.
  • And the boring stuff, which is what actually changes things: pay writers when you can, buy from small bookshops, ask your library to acquire titles, back open-access mandates and text-and-data-mining exceptions in law. Rebuilding public access is slower than downloading, but no judge prosecutes you for it.

A library that closes with a letter was never a library

We live in the strange moment when a perfect copy of much of human knowledge exists, is searchable, weighs a petabyte, and cannot be removed without confiscating drives on half the planet. At the same time, the business complaining about it has managed to put two people at risk of twenty years in prison for uploading manuals, and has ended up buying that same copy wholesale to feed its models.

The useful question is no longer whether these libraries exist. They do. It’s whether we can build legal access good enough, cheap enough and universal enough that shadow libraries stop being the only way to read. Meanwhile, the market answers for us: in the shadows.


Written in November 2026. Figures and dates checked against each project’s public catalogue, US Department of Justice press releases, TorrentFreak, Publishers Weekly, The Atlantic, AP and RELX’s annual reports.

Publicado en: Blog

Tags: soberanía-tecnológica cultura-hacker open-access ia libros

Want to subscribe to new posts?