If a book is worth learning from, it is worth preserving.
Artificial intelligence companies need enormous amounts of human knowledge to train their models.

Increasingly, that knowledge is being found somewhere very old-fashioned: inside physical books.

Investigations and court records have revealed that millions of physical books have been purchased for AI data acquisition and subjected to a process known as destructive scanning.

The binding is removed.
The spine is cut away.
The pages are separated and scanned at high speed.
The physical book is then discarded or recycled.

For easily replaceable mass-market books, that is already wasteful.

For rare, antiquarian, out-of-print, locally published and culturally significant books, the consequences can be irreversible.

A book is more than the sequence of words printed inside it.

Physical books carry annotations, printing techniques, typography, illustrations, bindings, ownership marks, marginalia and evidence of the cultures that produced and preserved them.

Once destroyed, that physical record cannot simply be downloaded again.

AI should help humanity preserve and expand access to knowledge, not reduce the number of surviving objects that contain it.

We are therefore calling on Amazon, Anthropic and every company acquiring physical books for artificial-intelligence training to adopt a Preservation-First AI Training Standard.

We ask them to:

1. Never destructively scan rare, antiquarian, limited, out-of-print or otherwise scarce books.
2. Use non-destructive digitisation whenever a book may have historical, cultural or archival value.
3. Establish safeguards to identify potentially scarce books before destructive scanning occurs.
4. Preserve or donate physical books after scanning whenever preservation is possible, working with libraries, universities, archives and independent booksellers.
5. Require contractors and data suppliers to follow the same preservation standards.
6. Publish transparent policies explaining how physical books are acquired, evaluated, scanned, preserved and disposed of for AI training.
7. Allow booksellers and marketplaces to know when bulk purchases are intended for destructive digitisation and give sellers the ability to opt out.
8. Support shared, preservation-first digitisation initiatives so the same physical works do not need to be repeatedly purchased and destroyed by competing AI companies.

This does not require stopping AI companies from learning from books. It requires something much simpler:

Scan the knowledge. Save the book.

Technology and cultural preservation should not be opposing forces.

The companies building the most advanced knowledge systems in history should also be capable of protecting the physical knowledge that made those systems possible.

Sign this petition and ask the AI industry to make one simple commitment:

Train AI. Preserve books. Save the source.