I'm generously excerpting this piece for you because it's behind the paywall of a professional publishing journal, and because I think it reveals just what a fraught, uncertain, unregulated, dangerous place the work of writing and publishing has arrived at.

Full credit to Erin Somers, and link is here: lunch.publishersmarketplace.com/2026/07/seven-figure-book-deal-cancelled-over-ai-suspicions-raise...

**
Seven-Figure Book Deal Cancelled Over AI Suspicions Raises Questions About AI Guardrails

By Erin Somers

A debut thriller won by Minotaur in a seven-figure, two-book deal after a reported 14-way auction was withdrawn by its agents this week for suspected use of AI. The novel, whose title and author PL is withholding for the author's privacy, sold for an alleged $2.5 million or thereabouts. But rumors surfaced quickly from people who had read the manuscript that the book contained many of the hallmarks of AI, including negative parallelism, off-kilter metaphors, and a flat prose style. On Wednesday, the author's co-agents Marc Gerald and Ashley Coleman at Europa Content withdrew the book, after questioning the author about the validity of the rumors.

St. Martin's Publishing Group would not confirm the size of the deal, telling Publishers Lunch only, "Macmillan Publishers participated in an auction. The agent ultimately pulled the book."

Europa wrote in a statement that the author, "Produced an incredible manuscript that dazzled us and the top publishing professionals around the world."

[Inserts question: How did the off-kilter metaphors and flat prose style dazzle all those top publishing professionals? But I digress. The report continues:]

"Unfortunately, we are no longer able to authenticate how the manuscript fully evolved from origin to completion. During the submission process we vetted the manuscript... and felt trusting in [the author's] reassurance that it was written without utilizing AI as a resource in the writing or editorial process. We can no longer substantiate that, and we are withdrawing the book. This raises so many questions about authorship and what AI means for this industry but those are questions for another day. We regret that our enthusiasm led to this outcome and we apologize to our publishing partners."

Gerald further explained to PL that AI suspicions were first raised during the submission process by an editor who had read the manuscript: "We immediately had a very long, substantive call with [the author], about AI protocol and the consequences for both him and Europa if he was using it. We went through detailed sections of his book, where he shared with us that like many authors he used AI for research into chemistry which forms the background for much of the book. That said, he confirmed he wrote the book 100 percent himself. At no point did we run an AI detection on the MS; we had concerns that the software had lead to previous inaccuracies."

The author denied using AI to PL, writing, "The rumors are absurd." But this week, Gerald and Coleman spoke to the author again and told the Bookseller that "aspects of [the author’s] story changed."

Gerald wrote to PL: "On Wednesday, Ashley and I decided we had no choice but to pull the book after we weren’t able to confirm for ourselves that AI had only been used for light research, grammar checking and Americanizing the language. While we are not able to say authoritatively whether or not he used AI, we did have trouble validating how the work went from creation to polished draft. Ultimately, trust between an agent and client is everything - especially given the fact that AI detection software is still in its infancy."

An analysis by Pangram 4, the leading AI detection program, returned an assessment of 97% AI...But there has been a backlash to Pangram and competing programs, which some people dismiss as inaccurate. Reported individual tests of other material on previous versions have shown more idiosyncratic results and people have challenged the company's claims. Chakrabarty said, "There are little false positives here and there. And there is some acceptable margin of error. For instance if it says an article is 80-90% human you can assume that its predominantly human with some AI edits etc which is totally fine and not egregious. But when an 80K word + novel say it's 97% AI generated it's very hard to deny it." He added that AI hallmarks that editors can look for include: plot inconsistency, too many mixed metaphor cliches, too much negative parallelism, and rule of triad, or excessive groupings of three...

Some authors will use generative AI to "edit" rather than write, without necessarily realizing the extent to which the software will change text unless blocked by specific prompts. Chakrabarty notes, "It rewrites everything and changes a writer's voice. As Hemingway said, Writing is Rewriting. If ChatGPT does edit it's still AI writing."

Sources tell PL that more deals are being cancelled behind the scenes following the discovery or suspicion of AI -- one publisher noted that multiples imprint have had to cancel a book over this -- including books that originate in the self-publishing sphere and are then picked up by Big Five publishers. This usually happens quietly and under the public radar, and is only brought to light by fans online...

In addition to issues in traditional publishing, AI-generated books are now flooding the self-publishing market, cutting into the livelihoods of human authors. Chakrabarty recently led a study investigating the amount of AI in self-published KDP genre fiction, finding "our results bear directly on the market-effect question at the center of the fair use defense to copyright infringement."

...The authors concluded, "Generative AI can...reshape a creative market through scale rather than quality.... We call this dilution." They note, "Our findings reveal a market remade by scale. Books with substantial AI text enter the catalog far faster than the revenue available to absorb them, push down the return each title earns, and contest the top ranks, while the books with AI text that succeed draw most heavily on the distinctive language of existing books."

In their piece on the new research study, the NYT wrote, "The researchers used Pangram to analyze text from around 60,000 books sold on Amazon. From that sample, about 53 percent of the books published in 2025 contained A.I.-generated text, compared to 23 percent in 2023."

In another small but positive development, ChatGPT has stopped fulfilling prompts to generate text in the style of a specific writer. Ars Technica writes, "In refusing to directly copy the 'exact style' of various authors, ChatGPT offered instead to capture an overall 'feeling' by incorporating some of the common features found in those authors’ work."

Still, the vetting tools remain imperfect, and questions abound. The author of the withdrawn book told PL, "Writers are increasingly being made to second-guess ordinary stylistic choices because they might be labelled as signs of AI. Even em dashes are now treated with suspicion, which makes me wonder what people would say about Emily Dickinson’s writing today."

It's clear that this is a watershed moment in publishing and agents, editors, and authors will continue to flounder without firm procedures in place. The core problem remains: there is no way to tell for sure.

**

I don't have any wise words of summary, just a lingering sense of dread. ... See MoreSee Less

Seven-Figure Book Deal Cancelled Over AI Suspicions Raises Questions About AI Guardrails - Publishers Lunch

lunch.publishersmarketplace.com

A debut thriller won by Minotaur in a seven-figure, two-book deal after a reported 14-way auction was withdrawn by its agents this week for suspected use of AI. The novel, whose title and author PL is withholding for the author’s privacy, sold for an alleged $2.5 million or thereabouts. But rumors...

23 hours ago