{
  "id": 3125005,
  "title": "'Unlicensed, unrestricted AI training could destroy the ecosystem for books' — quote of the day by the Authors Guild on the sourcing of training data",
  "url": "https://urgent.news/2026/08/24/unlicensed-unrestricted-ai-training-could-destroy-the-ecosystem-for",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-24T22:00:00.000Z",
  "source": {
    "name": "TechRadar",
    "slug": "techradar",
    "url": "https://www.techradar.com/pro/unlicensed-unrestricted-ai-training-could-destroy-the-ecosystem-for-books-quote-of-the-day-by-the-authors-guild-on-the-sourcing-of-training-data"
  },
  "original_language": "en",
  "account": "Large language models, like the Claude AI developed by Anthropic, require substantial data for effective training. Companies have sourced this data from various origins, including publicly available internet content, AI-generated synthetic data, and printed literature. However, recent allegations have surfaced that some AI firms have sourced their training data from books available on pirate ebook sites, a practice that has drawn criticism from writers, authors, and publishers.\n\nThe Authors Guild, a professional organization for authors, has issued a comprehensive guide on AI licensing in response to these concerns. The guide covers the various ways AI companies use published works, including both legal and illegal methods. More worryingly, the guide warns that continued data harvesting could jeopardize the ecosystem for books as we know it.\n\nOne specific concern highlighted is the existence of Project Panama at Anthropic. This scheme involves the destructive scanning of all books in existence. The term \"destructive\" refers to the fact that the process would include scanning the books and then destroying them. This initiative is shrouded in secrecy, hence the codename, as Anthropic does not wish to publicize its efforts. The high quality of books compared to online writing is a factor that makes them an attractive source for AI training data.",
  "summary": "Many of today's widely used AI systems have been trained on material sourced from books",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}