On the surface, American book publishing has spent the past year behaving like one of the few industries willing to draw a line in the sand against the rise of generative AI. Headlines have trumpeted copyright lawsuits against major tech companies, the cancellation of books by authors suspected of using large language models, and the kind of high-minded language about creative integrity that makes readers feel that the people who bring us literature are still fundamentally human, still stubbornly protective of the written word. Yet the public battle is only half the story. Behind closed doors, according to more than two dozen publishing workers who spoke on condition of anonymity for fear of jeopardizing their employment, at least three of the Big Five American publishers — HarperCollins, Simon & Schuster, and Hachette — have been quietly integrating AI tools into their daily operations for months. They are using them to email literary agents, to draft publicity and back-cover copy, to design cover art, and to generate marketing campaigns, often at the explicit direction of senior executives. This has happened with little public disclosure and none of the consent of the authors whose work, in many cases, is the very material the AI is learning from and replacing. The result is a portrait of an industry at war with itself: fiercely defensive about the sanctity of books in the public square, yet increasingly dependent behind the scenes on the very technology it is publicly decrying.
The most striking thing about this quiet adoption is not that it is happening, but how deeply unglamorous it is. These are not futuristic projects conducted by experimental research departments; they are mundane tasks woven into the ordinary rhythm of publishing. Publicists use chatbots to draft emails to literary agents, editors use AI models to sharpen jacket copy, designers use image generators to develop cover concepts, and marketing teams use automated platforms to plan social media campaigns. The work is not necessarily brilliant, but it is fast, and in an industry perpetually praised for its romance and chronically short on time, speed has become its own justification. Several editorial staffers said the widely publicized cases of authors accused of using AI to help write their books — a horror novel from Hachette, a crime thriller from Macmillan — are merely the visible crest of a much larger wave. “Every single imprint has a story of someone very esteemed who had a book canceled because of blatant AI use,” one editor said, “and that story never made the news.” The implication is unsettling: the scandals readers see are exceptions that were caught, while a much broader culture of AI-assisted production is quietly becoming business as usual across all levels of the industry. The executives pushing these tools are not necessarily villains; in many cases, they are stressed decision-makers responding to shrinking margins and the relentless demand to publish more, faster. But the people tasked with making AI useful are often the same junior and mid-level employees who worry most about what it means for their jobs, their ethical commitments, and the authors they work with.
Nowhere is the contradiction more visible than at HarperCollins. According to a current employee, senior leaders purchased Claude licenses from Anthropic months ago “but very quickly realized they didn’t know how to implement them.” So, in a familiar corporate move, dozens of junior and senior staff members were “voluntold” to become “AI Champions.” Their task was to discover practical use cases for the technology, which they did in monthly brainstorming sessions with colleagues across departments. The title alone — “Champion” — suggests a kind of crusading energy, but the mood inside those meetings was often tense. This spring, when some staff members raised concerns about the legal, ethical, and environmental consequences of adopting AI, a person leading the meeting waved the objections away, dismissing them as “the cost of doing business.” The phrase landed like a small betrayal. Publishing has always sold itself as a human business, one where taste, empathy, and relationships matter as much as quarterly profits. To hear a leader describe the environmental and ethical damage of AI as an acceptable expense, rather than a deep existential challenge to the industry’s values, felt to many like an admission that the commercial machine had finally swallowed the literary mission. Shortly afterward, a statement appeared in the new AI section of the HarperCollins employee portal, acknowledging that “while AI does have an environmental impact, it’s a small part of most individuals’ total digital carbon footprint.” The defensive framing bothered staff members even more. It felt like an official corporate gloss designed to soothe consciences rather than to engage with the difficult truth: that data centers consume enormous amounts of energy, that creative labor is being devalued, and that none of this was discussed openly with the authors whose livelihoods depend on the integrity of the publishing process.
The resentment at HarperCollins is not only about ethics; it is also about money. Publishing has never been a lucrative industry, especially at the entry level. The average salary for New York City publishing employees at the Big Five and Scholastic was just $47,583 in 2023, a number that sounds all the more painful when set against the cost of living and the prestige expected from people who work with celebrated authors. One staffer described the arrival of AI as “a slap in the face.” “We’ve been making valid business cases for years to be paid a healthy salary,” they said, “and instead they spent that money on software that most of us don’t really want.” The complaint is not merely about compensation. It is about a disparity between corporate priorities and the people doing the day-to-day work. To some employees, investing in expensive AI subscriptions while failing to pay adequately the humans who actually edit, promote, and care for the books feels like a betrayal of the industry’s basic promise. And it is not just one tool. HarperCollins has purchased licenses for Claude, ChatGPT, and Jasper, an agentic marketing platform that claims to be able to run end-to-end marketing workflows. At least one division, according to one staffer, has gone “all in” on AI, producing AI-generated marketing videos and automating tasks that were once the province of human creativity. It is hard to imagine a more symbolic loss than the replacement of human-generated marketing content with machine output in a company whose entire purpose is to champion human expression. Yet that is exactly what is happening, one quiet department at a time, with no author consultation and no reader awareness.
At Simon & Schuster and Hachette, the picture is similar, though less centralized. Workers at both houses say their employers have been experimenting with large language models for some time, and that one of the most common use cases is generating publicity copy. This might sound innocuous — promotional language is hardly sacred — but in practice it is a profound shift. Publicity copy is the voice through which a book first introduces itself to the world. It is meant to capture the author’s tone, the book’s emotional register, and the singular vision that no machine could fully understand. When a blurb is generated by a model that has digested millions of other blurbs, it tends to flatten into a familiar, safe, and generic shape. It may not be bad; it is simply not human. The deeper, more troubling issue is that many authors have no idea their publishers are using AI at all. They submit their manuscripts, receive a jacket and a marketing plan, and assume that everything was made by the same editorial culture they chose to trust. The article’s anonymous sources described a growing unease among staffers who feel they have been placed in an impossible position: they are expected to use the technology, but they are not allowed to tell the public, and they are certainly not allowed to tell the authors. This secrecy cuts to the heart of what publishing has traditionally been: a relationship of trust, built over time, between writers and the houses that agree to carry their stories into the world. If that trust is broken by an unacknowledged reliance on algorithmic tools, the damage may be slow to surface, but it will be deep, because it touches the essential contract that makes publishing meaningful in the first place.
It would be easy to dismiss these concerns as the nostalgia of an industry reluctant to modernize. But the workers in the article are not Luddites; many are young, adaptable, and eager to use new technologies in ways that genuinely help. What they object to is not innovation but its secretive, top-down, and ethically muddled implementation. They want the people who own the companies to make choices with open eyes and open explanations. They want authors to have at least as much information as readers. And they want the industry to ask an honest question: if publishing begins to rely on the very machine-learning models built from the uncredited work of writers, what, exactly, does it have to offer writers anymore? The public lawsuits against Google and the cancellations of AI-tainted books suggest that publishers understand the danger in principle. The quiet adoption of AI into their own workflows suggests that, in practice, the pressure to cut costs and maximize speed has become too strong to resist. Perhaps the most honest summary of this strange moment is that publishing has not decided whether AI is an enemy or an ally; it is living with both at once, with no moral center, no transparency, and no consent. The industry that has spent years fighting for the human right to tell stories is now, behind the curtains, letting the machinery of algorithmic generation remake its own processes. If that continues without honest conversation, the biggest story publishing produces next may not be in any book at all. It will be the story of itself.
