2mo ago

Meta staff torrented nearly 82TB of pirated books for AI training — court records reveal copyright violations

www.tomshardware.com

Summary

Court records in an ongoing lawsuit reveal that Meta staff allegedly downloaded 81.7TB of pirated books from shadow libraries like Z-Library and LibGen to train its AI models.

Internal messages show employees raising ethical concerns, with one saying, “Torrenting from a corporate laptop doesn’t feel right.”

Meta reportedly took steps to hide the activity.

The case is part of a broader debate on AI data sourcing, with similar lawsuits against OpenAI and Nvidia.

Technology @beehaw.org

melp @beehaw.org

2mo ago

Meta staff torrented nearly 82TB of pirated books for AI training — court records reveal copyright violations

www.tomshardware.com /tech-industry/artificial-intelligence/meta-staff-torrented-nearly-82tb-of-pirated-books-for-ai-training-court-records-reveal-copyright-violations

47 comments

Odds of severe punishment? Slim.
- Word on the street is that they might be facing almost a dollar per petabyte.
  
  Street value $8.7 trillion
- Odds of ~~severe~~ punishment? ~~Slim~~ None.

Remember Aaron Swartz? Do you think Zuck will go to prison too?
- yes, he'll be tormented by the feds to the point he'll take his own life.
- lolno. Fuckerberg won't see the inside of a prison cell. He's what we like to call in the law industry, "rich"

Rules for thee, not for me.

I'm mostly upset that this puts z-library and libgen back high up on the anti-pirating enforcers radar
- It’s fucked these guys can pirate all this shit and make money off it. But if the masses access it, shut it the fuck down! Break encryption! Curb the laws! Penalize! Penalize! Penalize!

So the "don't be evil" crowd casually torrented 82TB of shadow library data through corporate hardware. Internal messages show researchers knew it crossed ethical lines, yet Zuck personally greenlit circumventing copyright. The cognitive dissonance of building AI empires on pirated foundations would be poetic if it weren't so predictably dystopian.
This isn't oversight—it's systemic rot. Fines become tax-deductible line items while lobbyists ensure regulatory capture. When your legal team costs more than the penalties, infringement transforms into R&D strategy. The only surprise is anyone still pretending capital understands "ethics" beyond PR gymnastics.
Meanwhile indie authors get demonetized for quoting haikus. But sure, let's investigate if open models borrowed a few ChatGPT outputs. Nothing accelerates innovation quite like megacorps rewriting IP law through sheer audacity.
- "Don't be evil" was Google, and they abandoned that motto a while ago in favor of "think of the shareholders!"
  
  As soon as they became a publicly tradable company they are obligated, by the Dodge v Ford ruling, to only maximize shareholder value. Being not evil isn't an option.
- Your comments are a little odd. Why do you do all the bold callouts?

"When they do it it's progress - when you do it it's piracy"

Seed it Mark you asshole!
- He can't, the seed is stored in what Mark lacks!
- The biggest hit and run in the history of torrenting.

That is an insane amount of data. I'm trying to fathom what 82TB of text files looks like and I can't.
- So... if we say every ebook is 10mb (that's well into the high end, only a few are that big)
  That's 8,589,934 10mb books.
  AI says the average public library in the USA has 116,481 items (but that includes all media formats), but if we go with that, then 82 TB is about 73.74 average sized libraries with no repeating content.
  
  NYPL has around 10 million books and an additional 10 million manuscripts in its collection. Over 54 million total articles for lending.
  Not the largest by far, but still mind boggling in size.
  To torrent and ingest something of that size is crazy.

Time for the ol' slap-on-the-wrist few million dollar settlement, or whatever amount Facebook makes in a day; if the courts even bother to function at this point.
- Make it a royalty, until the data stops being used keep on paying.
- Jammie Thomas had to pay over $9000 per song she shared on Kazaa and that was like 15 years ago. Inflation + millions of shares should mean billions of dollars owed to the publishers... Plus obviously deleting or forfeiting ownership of all the models trained on that data, naturally.

So long as we’re not just singling out Meta. They’ve all done it.
At least Meta, with its Llama model family, has enabled the open source LLM space to flourish (along with Mistral, AI2, Alibaba, Eleuther, and many others).
What-aboutism. I know. I’m okay with what’s happening here in the sense that in return we’ve gotten magical (compared to the SoTA five years ago) models with seemingly emergent reasoning capabilities and expertise in basically every domain. That’s huge, even if it’s started to feel normal.
The issue, of course, is creatives whose content was stolen now losing out on opportunities or revenue that they relied on, meaning fewer creatives in the future and more AI slop.
Not seeding is hilariously on-brand for Meta though. Maybe it’s the ‘possession < distribution’ defence?
- So long as we’re not just singling out Meta. They’ve all done it.
  They have to single out Meta for the narrative to work. Objectively, this is about major content owners, corporations, wanting a piece of something other people have created. That's a tough sell, so you have to spin a story.
  Not seeding is hilariously on-brand for Meta though. Maybe it’s the ‘possession < distribution’ defence?
  Sorta. AI training is clear-cut fair use, which is why you get manipulative stories like this one. What exactly do these out-of-context quotes say about the law? Nothing, but it serves the narrative.
  Actually seeding the content is problematic. If you knew that the downloaders had some legal purpose, that might work. But just sharing it is hard to justify.
  
  Yes, we know: the law is carefully crafted so that it's only illegal if WE do it.

That is like all of the books...
- Yeah, they're not exactly large in file size.

Perfect example of 'rules for thee but not for me'. Assholes have no issue throwing the book at individuals infringing on copyright, then will turn around and pull heinous shit like this. Heinous in their eyes mind you.
- Heinous when it doesn't benefit THEM

How are these LLMs not a poison pill already? Anything they generate has got to be such a mashup of permissive and protective licensing.

As much as I hate much of the news about AI, I love the dilemma this puts the copyright lawyers and tech bros in. Either they admit that the majority of copyright law enforcement is a joke and stifles innovation - or they admit the creation of AI using stolen works is standard practise and requires government intervention to get back on track.

I said this in another thread on this subject- this is a very clear violation of the Berne Convention and Meta could find itself in court all over the world because of this.
https://www.wipo.int/treaties/en/ip/berne/
Probably not, but I can hope.