Library and Archives Canada Stops Granting ISBNs to AI Generated Books
Library and Archives Canada said it will no longer grant ISBNs for books generated by artificial intelligence. The move does not ban AI-generated books in Canada. It does not stop someone from uploading a file, selling a PDF, or printing copies privately. But it does draw a clear line around what Canada’s national ISBN agency is willing to recognize through the standard identifier used across the book trade.
The new position is straightforward in its main effect: AI-generated books will no longer receive ISBNs from Library and Archives Canada.
That puts Canada among the jurisdictions taking a firmer stance on machine-produced publishing. It also signals that ISBNs are not just neutral serial numbers handed out to anything that looks like a book file. They are part of a publishing and bibliographic system that depends on trust.
The ISBN system depends on human accountability. A publisher applies for the number, attaches it to a publication, and becomes part of a chain of recordkeeping. Even when a book is bad, rushed, niche, or self-published, there is usually a human or publishing entity making decisions about what it is, why it exists, and how it should be described.
A clear signal from Canada’s national library
Library and Archives Canada’s ISBN decision is one of the clearest signs yet that AI-generated books are being treated as a category apart from ordinary publishing. It protects a key part of Canada’s book infrastructure from being treated as an unlimited intake system for machine-made content.
For human authors and responsible publishers, the message is direct: keep creating, keep good records, and be clear about how the work was made.
The grey area is human use of AI
The hardest cases will not be obvious spam books. They will be works made with a mix of human and machine input.
If interested, you can view Broken Keys Publishing's position on the use of AI here.
The problem is not only the presence of AI. Many writers use spelling tools, grammar checkers, transcription software, translation aids, research assistants, and generative tools in limited ways. The problem is the mass production of books where human authorship, editing, verification, and responsibility are thin or absent.
That distinction matters. Library and Archives Canada is not saying people cannot use AI to produce text. It is saying that books generated by AI do not qualify for one of the key identifiers used in Canadian publishing infrastructure.
A writer might use AI to brainstorm chapter titles. An editor might use it to identify awkward sentences. A historian might use it to summarize notes before writing original prose. A novelist might ask a tool for timeline checks or alternate scene ideas. Those uses are different from asking software to generate the book itself.
The key question is likely to be whether the book is AI-generated or human-authored with AI assistance.
That line can be difficult to draw. It may depend on the author’s process, the amount of machine-generated material, the degree of human editing, and whether a real person made creative and intellectual decisions throughout the work.
A useful way to think about the difference is this:
AI as a tool
AI as the producer
A person uses software to support research, editing, organization, or drafting, while keeping control over the argument, narrative, facts, voice, and final text.
Software generates the main substance of the book, and the human role is limited to prompting, lightly editing, packaging, or uploading.
Library and Archives Canada’s position will likely push publishers to be more careful with internal records. If a publisher applies for an ISBN, it should be able to explain who wrote the work, how AI was used, and where responsibility sits.
That does not mean every author needs to fear ordinary digital writing tools. It does mean vague claims will become less useful. “AI helped me” is not enough detail. “I wrote the manuscript, used a grammar tool for copy edits, and fact-checked the final text myself” is clearer.
Self-publishers will feel the change first
Medium to Large publishers already have editorial processes, contracts, editing stages, and rights checks. Self-publishers may feel the policy more directly, especially those experimenting with fast AI-assisted production.
Canada has a strong self-publishing community. Many independent authors use ISBNs properly and publish careful, original work. Free ISBN access through the Canadian ISBN Service has been valuable because it lets Canadian publishers participate in the book trade without buying identifiers from a private vendor.
This decision does not target self-publishing as a category. It targets AI-generated books. Still, self-publishers may face more questions when applying for ISBNs if their books show signs of automated production or if their publishing model is built around large numbers of generated titles.
Will the policy solve every AI publishing problem?
Refusing ISBNs is a strong signal, but it is not a complete solution. Can a checkbox tell you who wrote a book? Of course not. But it's a start in the right direction.
There are also fair concerns about overreach. Some writers with disabilities use AI tools as assistive technology. Some authors writing in a second language use software to improve clarity. Some Indigenous, immigrant, rural, or independent creators may rely on affordable tools because professional editorial support is expensive.
A fair system needs room for those realities. The issue should be the nature of the finished publication and the human responsibility behind it, not whether a writer touched a modern tool.
The policy also leaves practical questions for the publishing sector:
How will applicants define AI-generated content?
Will partial AI use require disclosure?
What evidence will be needed in unclear cases?
How will appeals or reviews work?
How will the rule apply to translations, summaries, workbooks, and public-domain adaptations?
How will libraries and retailers describe books that used AI in some way?
Those questions will matter more as tools improve and become part of everyday writing software. A hard rule can be clear in principle while still hard to apply in real publishing workflows.
Comments