AI companies are burning books, advocates complain to FTC
A coalition of 18 advocacy organisations has formally asked the US Federal Trade Commission to investigate AI companies for buying physical books, scanning them for use as AI training data, and then destroying the originals. The complaint, revealed on Friday, centres on practices by Anthropic and Amazon, and argues the destruction of these texts is anti-competitive, stripping rivals and the public of access to irreplaceable resources while allowing the wealthiest firms to become sole custodians of digitised human knowledge.
Details of Anthropic's operation, codenamed "Project Panama," emerged through document disclosures in the Bartz v. Anthropic PBC copyright case, with a 2024 internal memo describing it as an effort to "destructively scan all the books in the world" and instructing staff to keep it secret. Older books are prized because they predate AI-generated text contamination, and destroying them after scanning cuts storage costs; a US court previously accepted this scan-and-destroy approach as fair use in Anthropic's case, though a similar defence failed for the Internet Archive. Book-data broker ISBNdb has since distanced itself from the practice, while Amazon and Anthropic did not respond to requests for comment on the advocacy groups' letter.
- 18 groups urge FTC to probe AI firms scanning then destroying books
- Anthropic's secret "Project Panama" scanned books, then shredded them
- Advocates say this hoards knowledge and harms competition and public access
New here? Start with this
AI developers use large collections of text to train systems that generate and respond to language. Books can be valuable training material, especially older works that predate the spread of AI-generated writing.
Anthropic and Amazon are among the companies named in a complaint by advocacy groups, which says some firms buy printed books, scan them and destroy the copies. Anthropic’s scanning project became public through documents in a US copyright case; a court accepted its scan-and-destroy method as fair use in that case.
The dispute is about who can access books and their digital copies, and how copyright and competition rules apply when companies build AI systems. The groups argue that destroying originals could limit access for libraries, the public and other firms.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The advocates raise legitimate concerns about concentrated control of cultural knowledge. When only the wealthiest companies can systematically acquire, digitise and then destroy books globally, they monopolise access to training data whilst eliminating rivals' and the public's ability to consult original texts. This loss of irreplaceable cultural artefacts cannot be recovered, and it grants disproportionate power to a small number of corporations over humanity's shared intellectual heritage.
The case against
The companies argue they operate within fair use doctrine, already validated by courts for Anthropic's practices. Destroying physical property one lawfully owns raises no copyright violation; copyright protects against unauthorised copying, not what companies subsequently do with owned items. Companies have legitimate efficiency reasons to avoid perpetual storage costs, and restricting their access to training data would harm innovation and consumer benefits.