Skip to content
Submit incident
AII-1000-37099
ReportedReported by fewer than three independent publishers, with no official finding.

Anthropic's "Project Panama" destroys millions of books for AI training

Anthropic reportedly bought millions of out-of-print books starting in 2024, scanned them for AI training data, and destroyed them—a practice it code-named "Project Panama". The company relied on US fair use law to justify the legally purchased destruction, though the practice raised concerns among authors, publishers and cultural heritage advocates in multiple countries.

28 July 2026 · Anthropic · Intellectual property and appropriation
28 July 2026
AII-1000-370992 sources ↓
Quick facts
Category
Intellectual property and appropriation
Country
United States
Source
7sur7
Date
28 July 2026
Drafted by
Raidu Incident Detection

What happened

Starting in 2024, Anthropic began purchasing out-of-print books by the millions through automated intermediaries, code-named 'Project Panama'. Books were mechanically scanned page-by-page and then destroyed, with the paper recycled. The company conducted this as clean training data for Claude, its AI model, to avoid contamination from AI-generated content that now saturates the internet.

Anthropic relied on the 'fair use' provision of US copyright law, which permits AI training on legally purchased copyrighted works provided the original is no longer accessible to the public. The operation was conducted secretly due to reputational and legal risks in other jurisdictions.

Reported impact

Authors, publishers and librarians across Europe and Australia expressed concern about the scale of book destruction for AI training. Cultural heritage advocates warned about the permanent loss of sometimes rare and ancient books solely for AI purposes.

Index analysis

The practice reflects the tech industry's massive demand for uncontaminated training data. Big Tech companies seek books published before 2022 in physical format to avoid AI-generated content that has polluted online sources with fabricated citations, invented information and plagiarized material.

Destruction of the books makes legal pursuit more difficult, as no evidence of the original work remains after scanning. The method is legal in the United States but presents judicial risk in other countries.

Governance controls

Anthropic justified the practice under US fair use law, which permits training on protected works if legally obtained, useful for learning, and no longer publicly accessible. The secrecy surrounding Project Panama indicated awareness of reputational and legal exposure.

Classification

Category
Intellectual property and appropriation
Industry
Media, social platforms and marketing
Country
United States
Harm status
Actual harm
Intent
Intentional
Timing
Unknown
Harmed
Civil society
Failure modes
Training data misuse
AI systems
Claude

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

7sur7
Opération Panama: des millions de livres, parfois anciens, sont détruits pour nourrir les IA
30 July 2026
Yahoo actualités
Anthropic détruit en toute légalité les livres après les avoir scannés pour alimenter son IA
31 July 2026
Oct 2026Sony and Warner Sue Anthropic Over Unauthorized Training on Copyrighted MusicAllegedIntellectual property and appropriationУКРІНФОРМSep 2026Biologist alleges Claude AI used unpublished research in enzyme discoveryAllegedIntellectual property and appropriationAnadolu AjansıCumhuriyetHaberler+1Sep 2026US government accuses six Chinese AI firms of stealing US AI modelsConfirmedIntellectual property and appropriationダイヤモンド・オンライン毎日新聞東京新聞 TOKYO Web+1Sep 2026Authors dispute publishers' claims on Anthropic copyright settlement paymentsWell documentedIntellectual property and appropriationThe New York TimesTechCrunchYahoo! FinanceAug 2026AI Companies Accused of Destroying Books for Model TrainingAllegedIntellectual property and appropriationAxiosCBS NewsCommon Dreams+1