Report · torrentfreak.com
Ex-Meta Researcher Torrented Pirated Books, Deleted It 'Due to Litigation'
torrentfreak.com
Source published October 31, 2026
Summary: 2 min read
50/100
Summary
Former Meta researcher Stephen Roller testified in a deposition that he torrented The Pile, a dataset that includes the pirated Books3 collection of nearly 200,000 books, according to rightsholders suing Meta in California federal court. The plaintiffs say Roller admitted knowingly distributing the material and said Meta had copies of Books3 when he left its FAIR AI research group. Roller worked at Meta from 2017 to 2023 and later deleted his personal copy, saying litigation made it contentious to keep copies of The Pile, the plaintiffs state. Meta disputes that the testimony proves Roller downloaded the dataset for Meta or while acting for Meta. The company says Roller could not recall when he downloaded it and testified it might have been for himself or another employer. Meta also says the deleted copy was on Roller’s personal computer and was removed more than two years after he left because he no longer needed it. Meta admitted in early 2024 that it used portions of Books3 to train Llama 1 and Llama 2, without specifying how it obtained the dataset. A prior ruling found training Llama on books to be fair use, leaving alleged BitTorrent downloads as a central unresolved issue. The plaintiffs also seek records related to an internal “ML Books” folder, a possible employee book club, and missing torrent-server command histories. Entrepreneur Media’s case is scheduled for trial in spring 2027.
Key Takeaways
- Rightsholders claim former Meta researcher Stephen Roller testified that he torrented The Pile, which includes the Books3 dataset of nearly 200,000 books sourced from the Bibliotik private tracker.
- Meta says Roller did not establish that his torrent download was for Meta, at Meta, or during his employment there, and says his deleted copy was stored on his personal computer.
- Meta acknowledged in early 2024 that portions of Books3 were used to train Llama 1 and Llama 2, but did not disclose how it acquired the dataset.
- Plaintiffs are seeking further records about Meta’s torrenting activity, including server command histories, as well as an internal “ML Books” folder and possible use of shadow-library books by employees.
- Entrepreneur Media’s case against Meta is scheduled for trial in spring 2027, while related claims from other publishers and authors continue to grow.
What the discussion said
The thread mostly rejects the easy question of whether ordinary people have ever downloaded unauthorized media. Readers instead focus on the AI-industry double standard: a student copying one textbook can face legal risk, while a company can allegedly ingest a vast pirate library to build a commercially valuable model and treat any eventual penalty as a business expense. For them, the troubling AI issue is not only copyright doctrine but power: frontier-model training turns mass acquisition of books into an input cost, while authors and smaller users carry the enforcement risk. Several commenters distinguish the legality of training from the legality of obtaining the corpus. They note that a court has found Llama training on books to be fair use, but argue that this does not erase questions about torrenting, possible organizational knowledge, or deletion of material amid litigation. Others point out that even licensed books would not settle the broader objection that models can imitate authors’ voices, so buying data may solve procurement liability without solving the cultural or economic conflict. The conversation also treats missing logs and deleted copies as especially corrosive: if companies can discard damaging evidence, discovery becomes too weak to test whether AI labs followed the rules they publicly invoke.
Where opinion split
The sharp dispute is whether the scandal is ordinary piracy rhetoric being selectively aimed at AI, or evidence that model builders receive a separate legal standard. Defenders of the narrower view stress that training itself has already received a fair-use ruling and that licensing would not end objections to stylistic imitation. Critics answer that fair use for training does not legitimize allegedly unlawful acquisition or evidence deletion, particularly when a corporation can absorb penalties that would crush an individual.
Coverage & discussions
Hacker News · torrentfreak.com
Community discussions
Community Sentiment
Positives
- A prior fair-use ruling on training Llama with books gives model developers meaningful room to learn from published material rather than requiring permission for every analytical use.
- Some readers see the debate as proof that data licensing alone will not resolve AI’s deeper author-rights problem, because concerns about style imitation survive even when the source books are paid for.
- The discussion usefully separates the model-training question from the procurement question, avoiding the simplistic claim that every use of copyrighted text by an AI is automatically unlawful.
Concerns
- Readers see a brutal enforcement asymmetry when a corporation can allegedly torrent a library-scale training set while individuals face disproportionate risk for copying a single textbook.
- Using allegedly pirated books as a foundation for a lucrative model looks to critics like deliberate cost avoidance, with expected legal exposure priced below legitimate licensing.
- Deleting local copies or leaving contested logs unavailable during litigation alarms commenters because it can make AI-training provenance impossible to audit after the fact.
- Even a legal training ruling does not calm critics who fear models can reproduce the market value of authors’ distinctive voices without compensation or consent.
Related Articles
Zuckerberg 'Personally Authorized and Encouraged' Meta's Copyright Infringement
May 5, 2026
AI-Assisted Paper on Z-Library Cites a TorrentFreak Article That Doesn't Exist
Oct 24, 2026
Anthropic trained Claude on illegally obtained Stanford professors' books
Oct 27, 2026
Meta faces 'astronomical' consequences as legal fight reaches critical moment
Aug 17, 2026
Business accounts will consent to Meta scraping their websites Starting Oct. 30
Oct 15, 2026