Since 2025, novelist Coral Hart has published 370 novels under 21 pen names with the help of AI, earning a high six-figure sum. As cases like hers emerge, publishers are canceling deals and authors are being accused of using machine-generated text—sometimes after checks that can be wrong. The publishing industry has yet to establish common rules: it is debating disclosure of AI use, copyright, and how to handle accusations. Researchers estimate that about one in five recently self-published books contains at least 25% AI-generated text.
AI is changing the book market and complicating the fight over authorship
How AI entered self-publishing
Hart began by writing romance novels for Mills & Boon, a publisher of paperback love stories. In 2012, she tried self-publishing dark romance—a subgenre about complicated and toxic relationships that was gaining popularity online but attracting little interest from publishers. Three years later, she moved entirely to self-publishing.
On Kindle Unlimited, authors are paid for the pages readers finish. The algorithm favors writers who release a new book at least once a month. Hart knows authors who hire more than a dozen ghostwriters to keep up that pace. She can write 10,000 words a day herself, but AI helps her increase her output. She says she has created 370 novels under 21 pen names since 2025 and earned “a high six-figure sum.”
Frequent releases alone do not guarantee income: if readers quickly abandon a book, it is not worth an author publishing one after another. Hart says AI does not replace an understanding of craft, structure, and reader expectations. Instead, it cuts the time involved and helps overcome business obstacles.
Hart now runs courses on using AI to create books. Until the law requires authors to disclose its use, she advises them not to, because of the public reaction. Hart says an open admission can lead to abuse, professional ostracism, and death threats. After a New York Times interview earlier this year, her identity was revealed, and she began receiving threats. Hart attributes the hostility to fear that the technology will displace people in publishing.
Accusations and canceled deals
There is another reason for the resistance: large language models may have been trained on copyrighted books without writers’ permission. Last year, a class action involving about 500,000 works ended in a $1.5 billion settlement with Anthropic, which was accused of using books without authors’ consent. Other similar cases are ongoing.
A group of 70 well-known writers urged the five largest publishers—Penguin Random House, Hachette, HarperCollins, Simon & Schuster, and Macmillan—to promise never to publish machine-generated books. The publishers made no such promise.
One author’s agent described the atmosphere in traditional publishing as “the Wild West”: the technology is moving faster than the industry can agree on rules. Liz Gately, a New York-based book scout, says the situation is changing and there are no clear answers yet.
Against this backdrop, canceled deals and public accusations have become more frequent.
Falade, like other authors the journalist contacted about these cases, declined to comment under his own name. In an essay, he noted that texts by authors for whom English is not a first language, or whose speech seems culturally unfamiliar, are more likely to arouse suspicion. He also suggested that the accusations against him might be linked to racism, and wrote that he had not seen the same determination to investigate cases involving white authors.
All the writers at the center of these controversies are people of color. In May, however, white author Stephen Rosenbaum admitted that he had accidentally included fabricated quotations in his book “The Future of Truth: How AI Is Changing Reality.” He said he had made a mistake by using AI and disclosed it in the acknowledgments. Rosenbaum regretted what happened, but called it a warning about the risks of using AI in research and fact-checking. The book was not taken off sale.
Checks do not dispel suspicion
It is difficult for an author accused of using AI to prove otherwise: there is no definitive way to establish where a text came from. Literary agents and book scouts say most AI-generated submissions are easy to spot. But human judgment can also be wrong, and research shows that authors for whom English is not a first language and neurodivergent writers are more likely to face false accusations.
Academic studies put Pangram’s accuracy in identifying AI-generated text at at least 95%, with a false-positive rate close to 0%. These figures indicate a low risk of incorrectly labeling human writing as machine-generated, but they do not settle any individual accusation.
Hannah Bowman, chair of the AI special committee at the Association of American Literary Agents, says agents disagree about what to do next: some believe AI detection programs should be used routinely, while others prefer to rely on trust. Bowman is concerned that disputes often become public after checks by journalists or social media users. Authors, she says, should have a chance to present their side, and publishers and agents should discuss concerns with them through a fair process.
Liz Gately says many agents now advise authors not to submit manuscripts as PDFs and to save several successive drafts rather than overwrite a single file. That makes it easier to respond to possible accusations.
The anxiety has led to some absurd cases. One agent received a manuscript from a writer he had worked with for a long time and who opposed AI. He noticed words in the text that seemed suspiciously strange. The writer was horrified: she had not used AI, but was so afraid of being accused that she had tried to make her prose “not sound like AI.” The effort backfired.
What publishers can and cannot do
A separate question is whether books should be pulled after accusations. Falade’s book had attracted interest from several publishers and received an offer of a multimillion-dollar advance. Literary expert Tad McIlroy says that does not resolve the question of how the book was written, but it does raise the question of whether suspicion alone is enough to halt work on a book. Falade recently signed with a new agent.
McIlroy was surprised when Hachette canceled the release of “Shy Girl” and believes he influenced the decision. After learning about rumors online, he checked the book with Pangram and contacted the New York Times. McIlroy expected the publisher either to edit the text or explain how AI had been used. “Shy Girl” first became a successful self-published novel; rumors about AI emerged before Hachette bought the manuscript. The publisher edited the text extensively, so McIlroy finds it hard to believe that nobody inside the company discussed the suspicions. He also thinks the book is strong and does not understand why the publisher would abandon it as soon as the story reached the media.
Both AI advocates and opponents are calling on publishers to set clear rules. McIlroy urges them to explain in detail how they use the technology internally and what authors are allowed to do. He suspects publishers are silent either because they deny what is happening or because they fear alienating their most profitable writers, many of whom are outspokenly opposed to AI.
Literary agent Tess Callero of Europa Content considers AI-generated fiction a frightening prospect. In her view, publishers are waiting so they do not rule out the possibility of using the technology. If the five largest publishers said they would not accept AI-generated texts, their position would carry more weight. Their silence, Callero says, leaves open the question of whether they see these books as a future source of income.
US law adds to the uncertainty: for copyright purposes, a book must be created by a human, while AI-generated text is considered public domain. Publishers therefore cannot obtain exclusive rights to it and lose legal protection. US law allows copyright protection for works created with AI, but text produced by a machine can still become the subject of litigation, even if the Copyright Office has registered the work after requesting few details.
Natalie Lacocil, a literary agent and chair of the copyright committee at the Association of American Literary Agents, believes courts will eventually consider the principle of human authorship. At the same time, she doubts the requirement will change substantially: copyright should protect human creativity. But, she says, the law should also take account of writers who want to experiment creatively with AI.
Lacocil compares the current anxiety in publishing to the arrival of e-books. Many feared for the market then, but it did not disappear—though it changed forever.
Experimentation and the limits of transparency
In 2023, Stephen Marche wrote the crime novella “Death of an Author” with the help of three AI programs. According to available information, it was the first AI-created novel to receive a review in the New York Times. Copyright questions meant it could be released only as an audiobook. Marche has been experimenting with AI in fiction since the early 2020s.
He believes the publishing world has become more moralistic while continuing to face scandals. Authors who understand the technology’s capabilities and limitations, and do not try to use it to deceive, are being left behind, he says. Those who commit fraud, meanwhile, will find a way regardless. Marche would like to see new forms of fiction created with AI; similar experiments are already underway in the art world.
He sees full transparency as the way forward: works created with AI should be clearly labeled, as should those that are not, just as fiction and nonfiction are distinguished today. Even opponents of machine writing can acknowledge that AI sometimes produces expressive results, particularly when the text does not pretend to be human. In her essay “Ghosts,” Vauhini Vara uses generative AI to write about her sister’s death—a subject she finds difficult to write about on her own. The resulting text is a conversation between a person and a machine about the expressive possibilities and limits of both kinds of language.
Marche expects AI books to appear in major publishing houses in one form or another, increasing the volume of low-quality writing. But, he points out, culture is full of dull and derivative work even without AI. He recently watched a Hollywood action film so formulaic it could have been made by AI, though it was not. In Marche’s view, worthwhile work in the future will be created by both people and AI, amid a mass of mediocre writing from both.
Last month, a new prize was established in France for authors of fiction who work with AI: the Prix Pascal de Littérature IA. Some AI-assisted books may be so convincing that an attentive reader will not be able to tell whether a human wrote them.
In a study by Tuhin Chakrabarty of Stony Brook University that has not yet been peer-reviewed, experts—students in leading graduate creative-writing programs—compared excerpts by their peers with texts by models trained on the works of well-known writers and then fine-tuned. In a blind test, participants more often preferred the AI text. It is not known whether they would have made the same choice if they had known how the texts were written.
Some readers may be willing to read such books. But literary agents interviewed by the article’s author see it as their job to protect human creativity. Doug Wallace, managing director of the literary agency network Andrew Nurnberg Associates International, believes AI will not become a major problem if the industry does not allow it to.
The article’s author observes that writing without machine assistance is becoming increasingly difficult: acquaintances in other professions react almost with pity when he says he would not use AI. His reasons are contractual obligations, reputational risk, and confidence that he can offer more than a machine. Yet machine-generated text is appearing even where one might expect resistance—not just in manuscripts, but also in leading newspapers and literary magazines. While the article was being written, the Times Literary Supplement removed an article about AI and the crisis of authorship after accusations that it had been written with AI.
Dataist’s take: The publishing industry must protect copyright, support fair checks, and decide how to disclose AI use—all at once. With no shared rules, the consequences of accusations depend on publishers’ decisions, and a public examination of a text can affect an author’s career before they have a chance to respond. Convincing machine-generated writing is already holding its own in blind comparisons with human work, but how readers will respond when its origins are fully disclosed remains an open question.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X