AI Detection Is Not Proof: Are Authors Becoming Guilty Until Proven Human?

Publishing has spent much of this summer asking whether AI wrote the book. A more uncomfortable question is now emerging: what happens when an accusation alone can damage an author’s career?

On 23 August, fresh reporting highlighted authors losing or seeing deals withdrawn amid suspected AI use even where the evidence remains contested. That follows a string of high-profile publishing disputes and growing concern about the reliability of AI-detection tools. The industry has a genuine problem to solve, but treating detection scores as proof risks creating another one.

Quick Shortcuts

Latest reporting on disputed AI-authorship claims

Publishers Weekly: industry-wide solutions to AI authorship

Oxford University Press AI guidance

Join the 121 Minutes Universe

The accusation is becoming almost as powerful as the evidence

The current anxiety is understandable. Publishers buy intellectual property. Agents stake their reputations on writers. Readers increasingly want to know whether the work they are buying was actually created by a person. When the origin of a manuscript becomes uncertain, trust can collapse remarkably quickly.

But there is a dangerous leap between asking legitimate questions and assuming a detector can settle them. HarperCollins chief executive Brian Murray has already argued that there may be no purely technical solution and that authors, agents and publishers need shared best practices. That is a much healthier direction than outsourcing judgement to software.

Detection is not provenance

AI detectors attempt to infer how text was produced from characteristics of the finished text. Provenance asks a different question: how did this manuscript actually come into existence? Drafts, notes, tracked changes, research files, outlines and version histories can tell a story that a percentage score cannot.

That distinction matters because writing is messy. Authors rewrite. Editors reshape sentences. Grammar tools intervene. Dictation software converts speech into prose. Word processors suggest corrections. Increasingly, AI may also help with research, organisation or administrative work. A modern manuscript can pass through plenty of technology without that technology becoming its author.

AI-assisted is not the same as AI-authored

This is where I think the publishing conversation needs more precision. I write my books. The characters, worlds, plots, prose and creative decisions are mine. I also use artificial intelligence around my work for research support, visualisation, advertising, online content and some of the machinery involved in running an author business.

Those are not equivalent activities. Asking an AI system to generate a novel and presenting the result as your own work raises a fundamentally different authorship question from using technology to visualise a starship, research a subject, organise marketing material or help manage a website.

If publishing collapses all of those activities into the single label “uses AI”, it will create rules that are both unfair and almost impossible to apply consistently.

Publishing is moving towards disclosure

More formal publishing environments are already developing clearer distinctions. Oxford University Press requires responsible and transparent declarations of generative-AI use and keeps primary authorial responsibility with humans. Elsevier similarly requires authors to document significant AI use in manuscript preparation while exempting basic spelling, grammar and punctuation checks. Nature Computational Science recently stressed the same principles: transparency, accountability and human oversight.

The direction of travel therefore looks less like a blanket ban on technology and more like a demand for meaningful disclosure. For writers, that may ultimately be preferable. Clear rules allow an author to say what technology did and, just as importantly, what it did not do.

The practical habit authors should start now

Keep your creative trail. You do not need to turn writing into a forensic exercise, but preserving dated drafts, outlines, notes and version histories is sensible. It protects more than AI provenance: it documents the development of your intellectual property.

Publishers also need restraint. A writer should not have to prove innocence every time an algorithm dislikes a paragraph. Evidence of process should support trust, not become a presumption that every author is guilty until they produce a digital alibi.

The real question

The industry does need a way to distinguish human-authored work from undisclosed machine-generated material. But the answer cannot simply be another machine making the accusation.

So where should publishing draw the line: should authors be asked to disclose meaningful AI use and preserve a reasonable record of their creative process or does that risk turning every writer into a suspect who has to prove they are human?

About Rob Frankson

Rob Frankson is a science-fiction author and creator of the Near Galaxy Saga. Through 121 Minutes he writes about storytelling, publishing, creativity and the changing relationship between authors and artificial intelligence.

Stay with 121 Minutes

Join the 121 Minutes Universe for AI & The Author briefings, writing updates, original science fiction and author resources.

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.