Where Does Publishing’s A.I. Problem Leave Authors and Readers? Major publishing houses risk unwittingly putting out books generated with A.I. tools. Authors and readers are frustrated, nervous and grasping for solutions.
Where Does Publishing’s A.I. Problem Leave Authors and Readers? Major publishing houses risk unwittingly putting out books generated with A.I. tools. Authors and readers are frustrated, nervous and grasping for solutions.
>Last fall, Antonio Bricio, an engineering consultant who lives in Guadalajara, Mexico, finished a draft of his first novel, a science fiction thriller about a government conspiracy to bury the history of humanity’s first contact with alien refugees.
>After querying 20 literary agents and getting a string of rejections, he spent several months furiously revising it in hopes of one day landing a publisher.
>Now, Bricio worries that the already taxing process of getting a publishing deal as a debut author has become even more fraught. He fears that agents and publishers will avoid taking risks on unknown authors over concerns that they might have written the book using artificial intelligence.
>The panic and paranoia over A.I.-generated books exploded last month, when a major publisher, Hachette, decided to [cancel the release of a horror novel](https://www.nytimes.com/2026/03/19/books/shy-girl-book-ai.html?searchResultPosition=1), “Shy Girl,” by Mia Ballard, in the United States over evidence suggesting that it had been partly produced by A.I. Hachette also pulled the book in the United Kingdom, where it released “Shy Girl” last year after Ballard initially self-published it.
>When Bricio learned about the novel’s cancellation on social media, his stomach dropped. He said he does not use A.I. to write, except to occasionally translate a stray word or phrase from his native Spanish into English, in which he is also fluent, using the A.I. translation program [DeepL](https://www.deepl.com/en/translator). But he wondered what an A.I. detector would say about his work.
CrayonsTasteLikeSky on
I imagine people who are unknowingly reading AI books are people who read what their algorithm tells them to read.
No_Humor_7952 on
yeah, it’s a wild situation right now. authors need to be protected from this tech, and readers gotta know what they’re getting into. authenticity in writing matters way too much to let AI take over.
There is no future for AI generated novels in the publishing industry, because if there’s no copyright, what’s the point? A publisher isn’t going to bother with an AI generated novel, because there is no way they can make money off it. Now they might not be able to effectively filter all AI generated content, but at the very least in publishing contracts I expect publishers to demand authors reaffirm that the book is human written.
Of course, there will always be shovelware e-books and print on demand books, but that’s at the very bottom of the market, where like, it would cost you $2 in tokens to generate a novel, and if you sell 1 copy you’re profiting….
InertBorea on
Remember that everything said to an LLM is stored somewhere. One day it will be possible to scrape datacenters for research, and people will discover which authors were “brainstorming” entire passages of their books in an AI chatbot.
Entire careers will be destroyed.
longjumpingtote on
We did a test where I work trying to be able to identify these things. Some of the tools were pretty good. But LMMs are good too. We would feed the initial output into an LMM and say “rewrite this to remove any hallmark giveaways that it’s by AI.” Then we’d take that output and have a different LMM scan that. Each iteration the chances of it being caught went down at least 30% so by the third iteration there was little to no chance of it being caught. But by the fifth, more problems started to emerge and it was more easily caught. (The texts we tested were human generated, and we “enhanced” parts with AI, like having AI fill out longer descriptions or tighten up dialog.)
The bigger problem (for us) was that they all identified some of the human-written parts as AI. We don’t want to reject something because it’s false-flagged.
So what do we do? Have a screen recording for the entire writing process?
6 Comments
>Last fall, Antonio Bricio, an engineering consultant who lives in Guadalajara, Mexico, finished a draft of his first novel, a science fiction thriller about a government conspiracy to bury the history of humanity’s first contact with alien refugees.
>After querying 20 literary agents and getting a string of rejections, he spent several months furiously revising it in hopes of one day landing a publisher.
>Now, Bricio worries that the already taxing process of getting a publishing deal as a debut author has become even more fraught. He fears that agents and publishers will avoid taking risks on unknown authors over concerns that they might have written the book using artificial intelligence.
>The panic and paranoia over A.I.-generated books exploded last month, when a major publisher, Hachette, decided to [cancel the release of a horror novel](https://www.nytimes.com/2026/03/19/books/shy-girl-book-ai.html?searchResultPosition=1), “Shy Girl,” by Mia Ballard, in the United States over evidence suggesting that it had been partly produced by A.I. Hachette also pulled the book in the United Kingdom, where it released “Shy Girl” last year after Ballard initially self-published it.
>When Bricio learned about the novel’s cancellation on social media, his stomach dropped. He said he does not use A.I. to write, except to occasionally translate a stray word or phrase from his native Spanish into English, in which he is also fluent, using the A.I. translation program [DeepL](https://www.deepl.com/en/translator). But he wondered what an A.I. detector would say about his work.
I imagine people who are unknowingly reading AI books are people who read what their algorithm tells them to read.
yeah, it’s a wild situation right now. authors need to be protected from this tech, and readers gotta know what they’re getting into. authenticity in writing matters way too much to let AI take over.
In the US, last month the courts have reaffirmed that copyright only applies to works created by humans: [AI Art Remains Ineligible for Copyright As Supreme Court Declines to Hear Case | PCMag](https://www.pcmag.com/news/ai-art-remains-ineligible-for-copyright-as-supreme-court-declines-to-hear)
There is no future for AI generated novels in the publishing industry, because if there’s no copyright, what’s the point? A publisher isn’t going to bother with an AI generated novel, because there is no way they can make money off it. Now they might not be able to effectively filter all AI generated content, but at the very least in publishing contracts I expect publishers to demand authors reaffirm that the book is human written.
Of course, there will always be shovelware e-books and print on demand books, but that’s at the very bottom of the market, where like, it would cost you $2 in tokens to generate a novel, and if you sell 1 copy you’re profiting….
Remember that everything said to an LLM is stored somewhere. One day it will be possible to scrape datacenters for research, and people will discover which authors were “brainstorming” entire passages of their books in an AI chatbot.
Entire careers will be destroyed.
We did a test where I work trying to be able to identify these things. Some of the tools were pretty good. But LMMs are good too. We would feed the initial output into an LMM and say “rewrite this to remove any hallmark giveaways that it’s by AI.” Then we’d take that output and have a different LMM scan that. Each iteration the chances of it being caught went down at least 30% so by the third iteration there was little to no chance of it being caught. But by the fifth, more problems started to emerge and it was more easily caught. (The texts we tested were human generated, and we “enhanced” parts with AI, like having AI fill out longer descriptions or tighten up dialog.)
The bigger problem (for us) was that they all identified some of the human-written parts as AI. We don’t want to reject something because it’s false-flagged.
So what do we do? Have a screen recording for the entire writing process?