.png)
Publishing’s AI Secrecy Problem: Why Writers Need Rules They Can Actually Use
There is a strange contradiction developing in publishing. The industry increasingly says it wants honesty about artificial intelligence, yet the consequences of admitting to almost any AI involvement can be so unpredictable that writers are being given a powerful incentive to keep quiet.
That tension is the subject of a timely argument in The Bookseller this week: if publishing genuinely wants transparency, it has to make honesty safer than concealment. I think that gets to the heart of the problem. We are trying to write rules for a technology that can be used in radically different ways, while too often treating every use as though it were the same thing.
In this article
Why fear is pushing AI use underground
The difference between AI writing and AI-assisted author work
Why AI detection cannot carry the whole burden
A more useful disclosure rule for publishers and authors
The cost of making honesty dangerous
We have already seen how combustible this subject has become. Last week the Académie Goncourt removed Thélyson Orélien’s novel C’était ça ou mourir from its prize longlist following allegations involving AI-generated text and plagiarism. Orélien denied using AI. Whatever eventually emerges about that individual case, the wider lesson is uncomfortable: an accusation of AI use can now become a reputational event before the underlying facts are settled.
That matters because publishing cannot build a workable policy around suspicion. If the practical message to authors becomes “mention AI and you may trigger an investigation, lose a deal or be treated as less authentic”, then many writers will simply stop mentioning it. The result will not be an AI-free publishing industry. It will be an industry with poorer information about how AI is actually being used.
Writing with AI is not the same as letting AI write
This is where I think the debate repeatedly loses its way. There is an enormous difference between asking a machine to generate a novel and using AI around the edges of an author’s work. I have no interest in replacing the author. The voice, characters, world, emotional judgement and final words should belong to the writer. If a machine produces the story and the human becomes little more than its publisher, that is a fundamentally different creative proposition.
But authors are also small creative businesses. Research, organising notes, checking continuity, exploring a scientific idea, creating a visual reference, analysing marketing, preparing metadata or helping turn a finished piece into promotional material are not the same act as surrendering authorship. Treating them as identical makes sensible disclosure almost impossible.
Jane Friedman’s updated AI and Publishing FAQ is useful here because it separates several issues that social-media arguments often collapse together: publisher practice, disclosure, human-authorship certification, copyright and AI training. The legal and contractual picture is still developing, which is precisely why authors need specific rules rather than moral panic.
Detection is evidence, not a verdict
There is another reason publishers need care: AI-detection systems are not magical truth machines. The Goncourt controversy itself has intensified debate about how much weight should be placed on detection scores, particularly when authors deny the allegation. A detector may be one signal worth investigating, but turning a probability score into a declaration of guilt is a dangerous shortcut.
The better question is not simply “Did AI touch this project?” It is “Who made the creative decisions, who wrote the work being sold as authored text, and what role did the technology actually play?” Those questions are harder, but they are also much closer to what readers care about.
A rule writers could actually live with
My preference is for a simple principle: disclose material AI generation of the work itself; do not force authors to catalogue every ordinary AI-assisted business or research task. If AI has written passages that remain in the published book, that is material. If an author has used AI to brainstorm a marketing plan, organise research, visualise a location or interrogate a continuity problem before doing the writing themselves, that is a tool supporting the author rather than replacing them.
Publishers can then add more specific contractual requirements where necessary. What matters is that those requirements are clear before the manuscript is submitted, proportionate to the risk, and consistent enough that an author does not have to guess whether admitting to a research assistant or image tool will suddenly place their career under suspicion.
Transparency only works when people know what they are being asked to disclose. Ambiguity produces fear, and fear produces secrecy. If publishing wants a culture in which writers are open about AI, it has to stop treating “AI use” as one undifferentiated act.
The human voice is still the point
For me, the line remains straightforward. AI can help an author work. It should not quietly become the author. That distinction protects the thing readers are actually buying when they pick up a novel with a person’s name on the cover: not merely sentences, but judgement, experience, imagination and a recognisable human voice.
The publishing industry now has an opportunity to build rules around that distinction. If it does, disclosure becomes useful information. If it does not, we risk creating exactly the culture nobody wants: one where writers who use AI responsibly feel safer saying nothing at all.
What do you think?
Should publishers require disclosure only when AI-generated text appears in the finished book, or should authors have to declare any AI use anywhere in the creative process?
About Rob Frankson
Rob Frankson is the author behind 121 Minutes and the Near Galaxy Saga, writing original science fiction while exploring how emerging technology is changing the practical business of being an author. About Rob Frankson
AI & Editorial Transparency
AI & The Author is edited and published by Rob Frankson. Artificial intelligence is used to assist with news research, initial drafting, content organisation and supporting imagery. All articles are reviewed and, where necessary, edited by Rob Frankson before publication. The opinions, editorial position and final decision to publish remain the author's.
Comments