The old fear was the blank page. A writer sat with an empty document and dreaded having nothing to say. The new fear is stranger and harder to dispel, because it arrives after the work is finished, after it is good, sometimes after it has won. It is the fear that someone will read your best sentence and decide a machine wrote it, and that you will have no way to prove otherwise.

The Minotaur case is the clearest example yet of how little proof the new anxiety requires. The debut, by Nigerian-born author Jerry Falade, had passed through the hands of fourteen editors during a bidding war, none of whom baulked. Then rumours began circulating that the manuscript carried what readers called the hallmarks of AI – a flat prose style, odd metaphors, a particular kind of repetition.

His co-agents withdrew the book. Notably, they did not claim to have established that AI was used. They said they could no longer authenticate how it had been written, which is a different and more slippery charge. No detection software was run. The decision cost the agency a commission worth several hundred thousand dollars, which suggests they did not take it lightly, and it also suggests something colder: that in the current climate, mere suspicion is now expensive enough to act on.

Falade has denied the accusations and hired legal counsel. He has posted texts and emails about the book dating to 2021, before ChatGPT existed. He has also placed the episode in a pattern that deserves attention, arguing that three Black authors received major deals in the same year and all three saw those deals cancelled or disrupted after AI suspicions, while white writers who have openly discussed using AI carry on unimpeded.

Literary prizes have become the second front in this, and the Commonwealth Short Story Prize has been caught there twice. What happened to two of its winners shows how little the accusation depends on evidence, and how much damage it does regardless.

Chanel Sutherland, a Vincentian-Canadian writer, won the 2025 overall prize for ‘Descend’, a story told in the voices of enslaved Africans aboard a sinking ship. Roughly a year later, someone ran the story through a detection tool, which returned a figure claiming the text was 88 per cent AI-generated. Sutherland denied it flatly. She attributed the repetition and cadence the tool had flagged to Vincentian oral storytelling, the tradition she grew up inside, where rhythm and refrain are the point rather than an error. Her publisher and agent stood by her, and no public review of the story’s origins has contradicted her.

Consider what the accusation asked her to accept. The very qualities that made her prose distinctive, its closeness to a spoken Caribbean voice, were being read by a machine as evidence that no human had been involved. A style rooted in a specific culture was flagged as the absence of a person.

The following year, the same prize faced fresh claims when its Caribbean regional winner, Jamir Nazir, was accused online of using ChatGPT. The Commonwealth Foundation stood by the result and made a decision worth noting. It stated plainly that it does not use AI detection software, because the prize is for unpublished fiction and feeding an author’s original work to a detector raises its own problems of consent and ownership. The tools, the Foundation added, are neither unfailing nor infallible.

That last point is the one the whole panic keeps colliding with. The detectors do not work reliably.

The case that should trouble everyone is not the guilty one but the false positive. A detector confronted with highly polished human prose produces exactly the same verdict it produces for a chatbot, because both are smooth, both are fluent, both avoid the small errors that mark a first draft. A writer’s competence has become indistinguishable, to a machine, from a machine. Sutherland’s oral cadences, Falade’s clean thriller prose, the well-made sentence in general, all of it now reads as suspicious to a system trained to distrust fluency.

Publishing has no defence prepared for this. As the industry has quietly acknowledged, there is no fact-checking apparatus in book publishing and no uniform policy for assessing whether a manuscript was written by a person. The checks come down to an editor’s instinct, which is precisely what the fourteen Minotaur editors were relying on when they bid.

The practical fallout is a change in what it means to be an author, and it is not a welcome one. A finished manuscript no longer speaks for itself.

Writers are now advised to keep everything. Every draft, every dated note, every version history, every scrap that might one day prove a human sat there doing the work. The manuscript is no longer sufficient evidence of its own creation. A writer must now also assemble a case file for the defence, against a charge that may never come, that cannot be disproved by the finished book, and that a flawed algorithm can bring at any moment.

The blank page at least belonged to the writer. It was a private struggle with a private resolution, and finishing the book ended it. This new anxiety has no such endpoint. A book can be completed, sold, published and garlanded, and still, years later, be dragged back into doubt by a number from a tool nobody fully trusts.

The writers most exposed are the ones with the most distinctive voices, because distinctiveness reads as anomaly. The safest way to survive the era, then, may be to write blandly enough that no machine ever mistakes you for one. That would be the worst outcome of all.