Ex-Meta Researcher Admits Under Oath He Torrented Pirated Books3
Deposition testimony contradicts Meta's claim that it never obtained the pirated book dataset via BitTorrent.
What to know
- Deposition testimony from former Meta researcher Stephen Roller contradicts Meta's prior denial that it obtained Books3 via BitTorrent.
- A court has already ruled Llama's book training is fair use, making how Meta acquired the pirated dataset the remaining contested issue.
- Plaintiffs say Roller's testimony reveals Meta had multiple copies of Books3, at least one torrented, and plan to seek more disclosure from Meta.
- Meta disputes the inference, saying Roller never confirmed the torrented copy was downloaded for or used by Meta.
The dispute Whether this is primarily a case of serious corporate/criminal misconduct worth prosecuting, or simply an unequal-enforcement issue where the underlying piracy itself is treated as unremarkable. · positions read across 6 posts and comments
Meta's conduct amounts to organized, potentially criminal wrongdoing, including destruction of evidence.
-
“This is organized crime, because the organization knew (and depended on the fact) that he was doing this, and it is also the intentional destruction of evidence.”
pessimizer · Hacker News ↗
The real issue is a double standard: ordinary pirates face lawsuits while Meta profits despite the same conduct.
-
“A student who torrents a textbook risks a lawsuit. Meta, reportedly, torrented hundreds of thousands of books, had an employee delete his copy “due to all this litigation,” and is now arguing over which server logs it has to hand over.”
AbhiDunna · Hacker News ↗
This fits a pattern of Meta employees misusing company infrastructure for piracy, not an isolated incident.
-
“meta has issues with employees using its network to pirate things, or its just a good time to try to eat with the AI scraper assumption”
someothherguyy · Hacker News ↗
Even if Meta had licensed the books legitimately, authors' objections to AI training on their work would persist.
-
“Even if they had bought all the books, wouldn’t people still be mad they trained an AI with it? I doubt it would quell the “It can emulate an Author’s style” complaints.”
0x20cowboy · Hacker News ↗
Stephen Roller Former Meta FAIR researcherMeta Defendant, AI developerVince Chhabria U.S. federal judgeSarah Silverman Plaintiff authorRightsholders/plaintiffs' counsel Plaintiffs in the consolidated lawsuits
How it unfolded 1 development · click the chart to see its coverage articlespostscomments
-
1
Story surfaces publicly, draws HN debate
TorrentFreak's report on the deposition spread to Hacker News and the fediverse, prompting debate over corporate liability, evidence destruction, and double standards in piracy enforcement.
“[d]ue to all this litigation…”
— Stephen Roller, Former Meta researcher · source -
This is organized crime, because the organization knew (and depended on the fact) that he was doing this, and it is also the intentional destruction of evidence.The fact that he was seeding a torrent is the thing that least matters here. They weren't making billions from seeding a torrent, they were spending bandwidth to do it. The fact that they…
2 more of the top 3 · 6 posts in this stretch
-
Everyone is arguing about whether piracy is good or bad, but that’s not what this case is about. It’s about whether the rules apply equally. A student who torrents a textbook risks a lawsuit. Meta, reportedly, torrented hundreds of thousands of books, had an employee delete his copy “due to all this litigation,” and is now arguing over which…
-
It’s interesting that it’s supposed to be very bad to destroy evidence, especially after litigation has started. But in the end, usually the consequences are better for the person destroying the evidence, than actually preserving itIf the evidence was very damning, and it’s not there anymore, the judge/jury can’t really properly make a…
-
-
background
Meta disputes the deposition's implications — Meta argued the status report misreads Roller's testimony, saying he did not know when he downloaded the dataset or confirm the download was for Meta rather than himself or another employer.
-
background
Plaintiffs reveal testimony, say Meta had multiple copies — Rightsholders filed a joint status report saying Roller's testimony shows Meta had multiple copies of Books3, at least one torrented, and that they will seek details on how many copies Meta acquired and how.
-
background
Roller testifies he torrented and distributed Books3 — In a deposition, former Meta FAIR researcher Stephen Roller, who worked at Meta from 2017 to 2023, admitted he downloaded Books3 via BitTorrent and knowingly distributed it, and said he deleted his copy about a year ago.
-
background
Judge orders Meta to name torrenting approvers — Earlier this month, the court ordered Meta to identify the employees who approved torrenting books from shadow libraries for AI training.
-
background
Meta denies torrenting Books3 dataset — Meta told Entrepreneur it did not believe it had obtained the pirated Books3 dataset, nearly 200,000 books taken from the Bibliotik tracker, through BitTorrent.
-
background
Judge rules Llama book training is fair use — A California federal court ruled last year that training Llama on books constituted fair use, leaving Meta's BitTorrent downloads of the dataset as the main open issue in the authors' lawsuits.
What people are saying 3 voices from 1 site · best of 6 · verbatim
- How many copies of Books3 did Meta actually acquire, and were any obtained officially versus by individual employees?
- Will courts treat Roller's deletion of his copy as spoliation of evidence?
- Oct 31
-
I don't know why, but piracy seems awful to me for things like books or music when it's one or a small group of people producing the media, but it doesn't seem too bad for films/series where they're often hard to get legally, and are created by large groups of people. It's not like any less human effort goes into them, it's just something I've…
-
Even if they had bought all the books, wouldn’t people still be mad they trained an AI with it? I doubt it would quell the “It can emulate an Author’s style” complaints.
-
meta has issues with employees using its network to pirate things, or its just a good time to try to eat with the AI scraper assumption