Zuckerberg 'Personally Authorized and Encouraged' Meta's Copyright Infringement
Publishers are suing Meta and Mark Zuckerberg, alleging the company torrented tens of terabytes of pirated books and articles to train its Llama AI models, raising the stakes in the fight over how copyright law applies to AI. Commenters contrast the aggressive prosecution of individual “pirates” like Aaron Swartz with the likelihood that a tech giant will face only fines, if any, despite allegedly infringing on millions of works. The exchange highlights deeper tensions over fair use, corporate impunity, and whether this moment should be used to tighten or dismantle existing copyright regimes.
Alleged conduct and scale
- Commenters focus on claims that millions of copyrighted books and articles were torrented (tens of terabytes) and used to train large language models, including seeding torrents and stripping copyright metadata.
- Some compare potential damages to a prior Anthropic settlement (billions of dollars; thousands per infringed work). Others note Meta’s huge cash and market value, arguing even multi‑billion fines might be a “slap on the wrist.”
Punishment, inequality, and corporate liability
- Strong sentiment that ordinary people and small-time “pirates” faced ruinous lawsuits and even prison, while large tech firms and executives expect at most monetary settlements.
- Many call for personal consequences (including jail) for executives who directed infringement, not just corporate fines, and criticize limited liability and current corporate personhood.
- Others argue criminal penalties for copyright are themselves unjust, but still insist that, until laws change, they should be applied equally to powerful actors.
Legality of AI training vs piracy
- One camp: training on copyrighted works is “like reading,” inherently transformative and fair use; the real legal problem is unlicensed acquisition (torrenting, ignoring robots.txt, piracy).
- Another camp: training necessarily involves copying; scale and commercial substitution matter, and models that regurgitate or function as market substitutes cross into infringement.
- Disagreement over whether current copyright law distinguishes humans vs machines, and whether scale (millions of works, industrial use) should change the analysis.
Views on copyright itself
- Several participants are anti‑copyright or “copyright abolitionist,” but still want laws enforced uniformly against big AI companies, partly as poetic justice for past harsh enforcement against individuals.
- Others favor reform: stronger protection for individual creators, weaker power for large rights-holders, and clearer rules for AI training and sampling‑like uses.
Enforcement mechanisms and precedent
- Some suggest RICO or criminal copyright statutes could apply if executives knowingly directed mass infringement. Others think the legal terrain around AI and fair use is still “clear as mud.”
- Comparisons recur to cases against file-sharing site operators, student MP3 downloaders, and high-profile prosecutions over academic journal downloads, highlighting perceived double standards.
Operational and ecosystem concerns
- Multiple reports of Meta’s crawlers aggressively scraping sites and ignoring robots.txt, prompting ASN‑level blocking.
- Open release of model weights is seen as making ongoing royalties for inference practically unenforceable, even if training is later found infringing.