Publishing Has an AI Trust Problem: When a Detector Accuses the Author

In this article

• Why AI-detection accusations are becoming a publishing problem

• Why detector scores are not the same thing as evidence of authorship

• What publishers and authors need from a fair disclosure process

• Why writers should retain control over legitimate AI-assisted workflows

A strange new problem is arriving at the writer’s desk. It is no longer enough to ask whether artificial intelligence can write a novel. Increasingly, authors may also have to prove that they did not ask it to write one.

This week that tension became unusually visible after Haitian-Canadian novelist Thélyson Orélien faced public accusations that his award-winning novel had been generated with AI. Orélien denied using AI to write the book, while his Canadian and French publishers initially stood behind him. Reuters reported the dispute as the novel continued to attract literary attention. Whatever ultimately emerges from the wider controversy, the episode exposes a problem the publishing industry has not properly solved: suspicion is becoming easier to create than certainty.

The detector is becoming the accusation

AI detectors are seductive because they appear to turn a complicated judgement into a number. Upload a manuscript, wait a few seconds, and a percentage appears. That looks scientific. But a probability score is not a chain of evidence showing how a manuscript was actually produced.

For writers, that distinction matters enormously. Style is not a fingerprint in the forensic sense. Clean prose, repeated sentence structures, formal phrasing or particular rhythms can be interpreted by detection systems as signs of machine generation even when a human wrote them. Editing can further complicate the picture. So can translation, grammar tools and the ordinary software assistance that has existed in word processors for years.

The risk is that the burden quietly reverses. Instead of somebody demonstrating that a writer used generative AI improperly, the writer is asked to demonstrate that they did not. That is a much harder proposition. How exactly do you prove the absence of a tool from months or years of creative work?

Publishing needs process, not panic

Publishers have every right to protect readers, copyright and the integrity of their lists. Authors also have a responsibility to be truthful about how a manuscript was created when a publisher’s rules require disclosure. But those principles do not justify treating an automated detector as a verdict.

A sensible process would separate evidence from indicators. A detector result might justify a conversation. It should not, on its own, establish authorship. Publishers can look at drafts, revision histories, editorial correspondence, research notes and the writer’s ability to discuss the development of the work. None of those is perfect individually, but together they provide something a percentage score cannot: context.

This is particularly important because the industry is simultaneously normalising other forms of AI use. Authors may use AI for research assistance, marketing, visualisation, transcription, accessibility or administrative work without handing over authorship of their fiction. Publishers themselves are exploring AI across workflows. A useful policy therefore has to distinguish between a machine producing the creative work and a writer using modern tools around a human-created work.

Disclosure has to work both ways

I have argued before that transparency is preferable to pretending these tools do not exist. I still think that. If a novel has substantially been generated by AI, readers and publishers have legitimate reasons to want that disclosed. But transparency cannot mean that human authors live under permanent algorithmic suspicion.

The Authors Guild’s model publishing clauses offer one useful direction. They treat AI rights and uses as things that should be expressly negotiated rather than silently assumed. That principle can travel further. Publishers should state what kinds of AI assistance they permit, what must be disclosed, what evidence they may use when concerns arise, and what opportunity an author has to respond before damaging conclusions are made.

Writers should also protect their own working trail. Keeping drafts, version histories and notes has always been useful. In an AI-suspicious environment it may become an additional layer of professional protection. I do not particularly like the idea that authors should have to archive their creative process as a defence file, but until publishing develops better standards it is a practical precaution.

The bigger question is trust

The argument about AI and books is often framed as human versus machine. I think the more immediate issue is trust: trust between writer and publisher, publisher and reader, and creator and technology company. Detection tools do not solve that problem. Used carelessly, they can make it worse.

A publishing culture that automatically distrusts every polished manuscript would be as unhealthy as one that ignores undisclosed machine-generated books. We need a middle ground built around evidence, clear contracts and proportionate disclosure. The objective should be to protect human authorship without turning authors into suspects simply because an algorithm dislikes the statistical shape of their sentences.

And perhaps that is the uncomfortable lesson from this week. AI does not only challenge the question of who can create. It is beginning to challenge how we prove that a human created something at all.

Sources & further reading

Reuters — Canadian and French publishers stand by author accused of using AI for prize-winning novel

Le Monde — AI and literature: the right of a writer to work with a machine

Authors Guild — AI-related model publishing contract clauses

About Rob Frankson

Rob Frankson is a science-fiction author and creator of the Near Galaxy Saga. He writes about storytelling, publishing, technology and the practical realities of being an independent author. About Rob Frankson

Stay with 121 Minutes

For more AI & The Author articles, science-fiction writing and publishing discussion, join 121 Minutes.

Join / Subscribe to 121 Minutes

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.