The AI book flood is real. The witch hunt it started is worse.
Two numbers describe the flood. Rakuten Kobo now rejects 45 percent of self-publishing submissions after a wave of AI-generated books, per The Bookseller. And Amazon's monthly e-book uploads have reportedly tripled to around 300,000 since ChatGPT arrived. The slop is real, it is industrial in scale, and platforms are right to build dams against it. Nobody serious disputes that part.
But watch what the panic is doing downstream. Granta stopped publishing Commonwealth Short Story Prize winners over AI accusations. The Commonwealth Foundation investigated the accused writers and cleared them. Granta withheld publication anyway. Read that sequence twice: an investigation happened, the writers were exonerated, and a prestige literary magazine still treated them as suspects. The follow-up coverage since June 20 has not made the decision look any better.
This is the part that needs saying plainly: suspicion is not evidence. AI detection is unreliable, the accusation is nearly unfalsifiable, and a writer with a clean formal inquiry behind them has done everything a writer can possibly do to prove authorship. When an institution's answer to exoneration is a shrug, that institution is running on vibes, and vibes are now overriding the verdicts of actual investigations. The scandal is not that readers got paranoid. The scandal is that the gatekeepers joined them.
A saner path appeared the same month. On June 29, Dutch publishers launched Bookpact.ai, an opt-in platform that licenses AI training rights per title, with consent and a price attached. Whatever its flaws turn out to be, it treats the problem as governance rather than contagion: decide what machines may touch, charge for the privilege, and keep human writers out of the crossfire.
The flood will keep rising and Kobo's filters will keep tightening, and both of those are fine. The line that has to hold is much simpler: cleared means cleared. A literary culture that will not honor its own investigations has a deeper authenticity problem than anything the bots are uploading.
