Anthropic's Book-Shredding for AI Training Sparks Copyright and Culture Debate
Recent court documents and media reports reveal that AI companies like Anthropic purchased millions of physical books, sliced off their spines for "destructive scanning," and destroyed the originals to build training corpora. This practice has sparked outrage over "AI book-burning" and cultural loss, but multiple analysts note this counterintuitive approach is an inevitable result of copyright law and judicial rulings, highlighting the deep conflict between AI training data acquisition and current legal frameworks.
Confirmed
- The case presents a counterintuitive legal split: Anthropic invoked the "first sale doctrine" to argue it had the right to destroy legally purchased books; however, 7 million books scraped from pirate libraries did not have this protection, resulting in a $1.5 billion settlement for that portion.
- A federal judge ruled that this practice of "completely destroying physical books while keeping only digital versions" constitutes "fair use" because destroying the original means there are no multiple copies. The judge's logic was that as long as the training process "transferred" text rather than "copied" it, it didn't constitute copyright infringement, though the legal consequences involved book destruction.
- According to reports from 404 Media and others, Anthropic spent tens of millions of dollars acquiring physical books, using hydraulic cutters to neatly slice the spines of used and even rare books for high-speed scanning, before sending the originals to recycling centers for destruction.
- Books published before 2022 became highly sought-after premium training data because they do not contain AI-generated text.
Unconfirmed
- Some sources claim AI companies are buying and destroying rare and out-of-print books, but counterarguments point out that 640,000 tons of books enter US landfills annually, many of which are merely outdated computer manuals, suggesting AI companies reusing them actually prevents resource waste.
Why it matters
- This case highlights the deep conflict between current AI training data acquisition and copyright law. As authors like @alejandroll10 and @iScienceLuvr pointed out, public anger might be misdirected: the real issue may not be the "scanning" or AI technology itself, but the legal clauses and judicial rulings that force books to be destroyed.
- @Kevin Bankston also believes the outrage over "destroying books" is somewhat hypocritical, as companies are merely adapting to the logic of infringement risk under recent court precedents, and the arguments criticizing companies today are somewhat the very forces that pushed the situation to this point.
- @aronchick and @heyabusiddik add a frequently overlooked perspective: hundreds of thousands of tons of discarded books are landfilled or pulped in the US annually, and using them for AI training might be a more rational form of reuse.
- @beenwrekt relayed Yoav Goldberg's criticism that existing copyright mechanisms force companies to destroy physical books after scanning and prevent them from sharing digital scans, an extreme restriction that makes direct piracy seem like a better alternative.
2026-07-29 ~ 2026-07-31 · 23 related posts
- Episode 1: AI Book Scanning and Destruction Sparks Copyright and Preservation Backlash(2026-07-27, 17 posts)
- Episode 2: AI Training Copyright Dispute Reflects Shift in Knowledge Access(2026-07-28, 2 posts)
- Episode 3: Anthropic's Book-Shredding for AI Training Sparks Copyright and Culture Debate(2026-07-29, 23 posts)
- Episode 4: Anthropic Exposed for Destroying Books to Train AI(2026-07-30, 4 posts)
- Episode 5: Anthropic Contractor Accused of Destroying Rare Books for AI Training(2026-08-02, 3 posts)
- Episode 6: Anthropic Accused of Destroying Physical Books for AI Training(2026-08-05, 2 posts)
- Episode 7: Booksellers Suspect AI Companies Destroying Rare Books for Training Data(2026-08-13, 2 posts)
Primary sources
- Anthropic Destructively Scanned Millions of Books to Legally Avoid Piracy Claims — soulbeddu ·
- Anthropic Reportedly Sliced Spines and Destroyed Millions of Physical Books for AI Training — 2C_ornot2C ·
- Anthropic’s book-copyright fight splits copying from destruction, with 7 million books and a $1.5B settlement — alex_verem ·
- Kevin Bankston says Anthropic’s book destruction follows the logic of recent copyright rulings — AndyMasley · 2026-07-29
- A pro-AI argument says 640,000 tons of books are landfilled every year — hey_abusiddik · 2026-07-29
- Hundreds of Thousands of Tons of Books Landfilled Annually: AI Training as a Better Alternative — aronchick · 2026-07-29
- A quote warns AI training may be destroying the original books it learns from — mtizard · 2026-07-29
- AI Companies Accused of Pulping Rare Books After Scanning for Training Data — heyshrutimishra · 2026-07-29
- Anthropic book-destruction criticism is really about copyright law and a judge’s order — alejandroll10 · 2026-07-29
- Anthropic copyright ruling sparks debate over book destruction and superintelligent lawyers — AndyMasley · 2026-07-29
- Post says the real problem in Anthropic’s book-scanning case was a judge’s destruction order — iScienceLuvr · 2026-07-29
- [source] Anthropic’s book-copyright fight splits copying from destruction, with 7 million books and a $1.5B settlement — alex_verem · 2026-07-29
- AI training consumes human civilization: Anthropic bought millions of physical books for training data — Graham_dePenros · 2026-07-29
- AI Companies Bulk-Buying and Shredding Rare Books for Training Data — alex_verem · 2026-07-29
- AI Companies Bulk-Buy and Destroy Physical Books to Scan for Training Data — emax · 2026-07-30
- Copyright Laws Backfire on AI: Destroy Scanned Books and No Sharing? — beenwrekt · 2026-07-30
- AI companies like Anthropic destroying books at massive scale, sparking data sourcing debate — SonnyDayCreates · 2026-07-30
- [source] Anthropic Reportedly Sliced Spines and Destroyed Millions of Physical Books for AI Training — 2C_ornot2C · 2026-07-30
- Anthropic Accused of Shredding Millions of Used Books for Training Data — l4rz · 2026-07-30
- AI Companies Reportedly Destroying Antique Books After Scanning for Training Data — moonsandhues · 2026-07-31
6 near-duplicate retellings: socialwithaayan · eyishazyer · heyshrutimishra · soulbeddu · heyshrutimishra · S_Conradi