In 2026, booksellers in several countries began reporting a strange kind of customer.
The orders were unusually large. They often contained hundreds or thousands of unrelated books: obscure academic works, old technical manuals, forgotten nonfiction, ordinary used books, and sometimes scarce titles. The buyers did not behave much like collectors, readers, libraries, or even conventional wholesalers.
Then the pattern acquired a possible explanation.
Court records in litigation involving Anthropic revealed that the AI company had spent millions of dollars buying millions of physical books. Contractors removed the bindings, cut the pages to size, scanned them, and discarded the paper originals. The resulting digital copies became part of a library used in developing large language models.
Booksellers began wondering whether some of the mysterious bulk orders they were receiving belonged to similar AI data-acquisition systems. Booksellers in Australia, Britain, Ireland, Canada, and the United States have since reported suspicious orders, although the ultimate buyer or purpose cannot be established in every case.
Some sellers welcomed a remarkable new source of demand.
Others refused.
At Maxwell’s House of Books in California, owner Craig Maxwell began cancelling orders he believed were associated with AI book brokers after learning about destructive scanning. He accepted that refusing the orders cost his small business thousands of dollars. His explanation was simple: he had entered the business to promote books and reading, not to participate in their destruction.
From a conventional market perspective, this may seem peculiar.
The bookseller owns a book.
A customer offers the asking price.
The bookseller sells it.
The buyer now owns the copy.
Transaction complete.
Pancakes sees something else.
A book is certainly a commodity.
It can be priced, bought, sold, shipped, inherited, donated, or discarded.
But commodity is only one relationship in which a book can participate.
A book can also be:
None of these meanings makes buying and selling books illegitimate.
They simply mean that the market transaction does not completely describe the object.
Pancakes frequently makes this distinction.
A thing may participate simultaneously in several relationships, and those relationships may give the same thing different meanings.
The AI acquisition process sees something like:
physical book
↓
text source
↓
digitization
↓
training corpus
↓
model capability
A bookseller may see:
physical book
↓
reader
↓
collection
↓
resale / donation / inheritance
↓
continued circulation
Neither description is imaginary.
The conflict appears because they imply very different futures for the same object.
Destructive scanning is not simply another buyer entering the used-book market.
It introduces a new transformation.
book
→ purchased
→ binding removed
→ pages separated
→ scanned
→ information retained
→ physical book discarded
The important institutional event occurs when booksellers realize what that transformation means.
Before learning about destructive scanning, an unusual bulk order might look like an unusually enthusiastic customer.
Afterward, the same order can be interpreted differently.
observation:
strange bulk orders
↓
new information:
AI companies are destructively scanning books
↓
new interpretation:
this transaction may permanently remove these copies from circulation
The seller has learned something about the system in which they are participating.
Pancakes treats this kind of change in perception as important.
Institutions are not static arrangements governed only by rules written in advance. Participants observe consequences, form better models of what is happening, and change their behaviour.
That is one mechanism of institutional evolution.
The booksellers who refuse these orders are also making an implicit distinction between ownership and stewardship.
Ownership answers questions such as:
Who may sell this copy?
Stewardship asks another:
What responsibilities arise from possessing it?
Those responsibilities are not necessarily legal obligations.
A bookstore owner cannot usually dictate what every future purchaser does with a mass-market paperback.
But booksellers can develop norms about their own participation.
A bookseller may decide:
I have the authority to sell this book.
That does not mean I am obligated to sell it
for every possible purpose.
The distinction becomes especially important as scarcity increases.
Destroying one copy of a recent bestseller with hundreds of thousands of surviving copies is not institutionally equivalent to destroying an unusual local history, an out-of-print technical text, a scarce edition, or a book whose surviving population is poorly understood.
A purely transactional model sees:
book → $40
A stewardship model may see:
one surviving copy
+
unknown future usefulness
+
limited replaceability
+
historical or scholarly value
The market price and the stewardship value need not be identical.
That does not mean every old book must be preserved forever.
It means destruction is itself a transformation with consequences worth noticing.
The booksellers’ response is interesting for another reason.
They are not withdrawing from the book market.
They are not arguing that books should never be sold.
They are saying:
We participate in this institution, but we will not participate in this particular transformation.
That is a form of institutional voice.
Pancakes treats participation as something richer than obedience.
A participant can cooperate, criticize, reinterpret, refuse, renegotiate, or leave.
Sometimes refusal is precisely how an institution discovers a boundary it did not previously know it needed.
The booksellers did not design rules for AI training twenty years ago because there was no practical reason to do so.
The technological capability arrived first.
Its institutional consequences became visible later.
The emerging sequence is therefore:
new technical capability
↓
new use for an existing resource
↓
unexpected demand
↓
previously invisible consequences
↓
participant concern
↓
local refusal
↓
discussion
↓
possible professional norms
That is institutional evolution happening in real time.
AI companies have an understandable objective.
High-quality books contain information that may improve models. Digitizing physical copies can also provide access to works poorly represented on the public web.
The technical question may therefore be straightforward:
Can we turn these books into useful machine-readable data?
The institutional question is different:
Under what conditions should we?
Pancakes separates capability from legitimate authority.
Being capable of performing a transformation does not determine whether everyone else must assist with it.
Likewise, ownership does not settle every normative question surrounding a resource.
The AI company may legitimately conclude that destroying a replaceable copy after scanning is acceptable.
A bookseller may legitimately conclude that participating in that destruction conflicts with the purpose of their own institution.
A library may reach a different conclusion.
An archive may distinguish disposable duplicates from preservation copies.
A collector may care about provenance.
An author may care about training rights.
A reader may simply want the book.
The institutional problem is not solved by pretending that only one of these relationships exists.
It is solved by making the relationships visible.
The dispute therefore turns on a deceptively simple question:
What exactly is a book?
For the scanning operation, the valuable resource may primarily be the information encoded in the pages.
Once that information has been captured, the paper object has completed its function.
For a bookseller, reader, archivist, collector, historian, or library, the informational content may be only part of what is valuable.
The physical object may possess continued capability:
book
→ read again
→ lent
→ resold
→ donated
→ compared with another edition
→ examined as an artifact
→ inherited
→ preserved
Destructive scanning chooses one capability and terminates the others.
That may sometimes be a reasonable trade.
But it is still a trade.
Pancakes would want the transformation represented honestly rather than disappearing behind the apparently neutral word purchase.
Independent booksellers occupy an unusual position in this story.
They were not appointed regulators of artificial intelligence.
They do not possess jurisdiction over the AI industry.
But they know their own domain.
They know what ordinary book orders look like.
They know which books are common and which are difficult to replace.
They recognize collectors, libraries, dealers, students, and habitual readers.
That local knowledge allowed them to notice an anomaly before they necessarily understood its cause.
In Pancakes terms, they became institutional sensors.
ordinary practice
↓
anomaly
↓
attention
↓
investigation
↓
new understanding
↓
changed practice
Their authority remains bounded.
A bookseller may refuse their own sale.
They cannot decide the fate of every copy in the world.
But bounded authority is still authority.
The refusal says:
Within the part of this system for which I am responsible, I will not help perform this transformation.
That is stewardship expressed through jurisdiction.
The interesting question is not whether every bookseller should adopt the same rule.
Probably they should not.
A warehouse full of unwanted duplicate textbooks presents a different stewardship problem from a scarce regional history. A publisher’s remainder stock differs from an antiquarian collection. Preservation scanning differs from training acquisition. Non-destructive digitization differs from cutting apart the only copy anyone knows exists.
The better institutional response will probably involve distinctions.
What is scarce?
What is replaceable?
What should be preserved?
Who should be consulted?
What information should buyers disclose?
When is destructive scanning reasonable?
Could non-destructive scanning accomplish the same purpose?
Should archives, booksellers, publishers, authors, and AI companies develop different practices for different categories of material?
These are governance questions produced by a new technological capability.
And that is why the booksellers’ refusals are more interesting than they first appear.
They are small acts inside a much larger process through which society discovers what a technology means.
A market description of the story is simple.
AI companies need books.
Booksellers have books.
Money changes hands.
Pancakes sees a more complicated institution.
Books participate simultaneously in markets, culture, scholarship, memory, and ordinary human relationships. AI training introduces a new transformation that preserves one aspect of the book—the information encoded in its text—while potentially destroying the physical capabilities attached to the copy itself.
Booksellers who refuse those transactions are therefore doing more than expressing affection for books.
They are making a stewardship judgment.
They have observed a previously unfamiliar consequence of their participation and decided that the ordinary rule—
customer pays
→ bookseller sells
—is insufficient to govern every new use of the resource.
Whether their judgment becomes a broader professional norm remains to be seen.
But the case illustrates something Pancakes considers fundamental:
Institutions discover their boundaries partly by encountering transformations they were never designed to govern.
Technology creates new capabilities.
Markets create new opportunities.
Neither automatically determines what people should participate in.
Sometimes institutional evolution begins when someone looks at a perfectly legal, profitable transaction and says:
No. That is not what I am here to help do.
That refusal is not outside the institution.
It is one of the ways institutions learn what they are for.
The existence and mechanics of Anthropic’s destructive-scanning program are unusually well documented. In Bartz v. Anthropic, the U.S. District Court described Anthropic purchasing millions of print books, having service providers remove their bindings and cut the pages for scanning, and discarding the paper originals afterward.
Reporting during 2026 documented unusual bulk orders received by booksellers in Australia, the United Kingdom, Ireland, Canada, and the United States. In many individual cases, booksellers could only suspect that AI acquisition was the ultimate purpose, so this article does not treat every unusual order as proven AI activity.
ABC 10News reported that Craig Maxwell of Maxwell’s House of Books in La Mesa, California, had begun cancelling orders he associated with AI book brokers after learning about destructive scanning, despite the financial cost to his independent store.
The institutional interpretation in this article—particularly the distinctions among ownership, stewardship, participation, resource meaning, and institutional evolution—is Pancakes analysis rather than a claim about the motivations of every bookseller or AI company.