At least three of America's five largest book publishers are deploying generative AI tools across their operations without public disclosure or author permission, according to a WIRED investigation published this month. HarperCollins, Simon & Schuster, and Hachette are using large language models like Claude and ChatGPT to draft emails to literary agents, compose publicity materials and back-cover descriptions, create cover art, and produce marketing content, often at executive direction. The investigation, based on interviews with more than two dozen publishing workers who spoke anonymously, reveals a gap between the industry's public stance against AI and its private embrace of the technology.

HarperCollins bought Claude licenses from Anthropic months ago and designated dozens of junior and senior employees as "AI Champions" tasked with finding use cases in monthly brainstorming meetings, according to current staff. When workers raised concerns about legal, ethical, and environmental consequences this spring, a meeting leader dismissed the objections as "the cost of doing business." The company has also purchased licenses for ChatGPT and Jasper, an agentic platform designed to automate marketing workflows. At Simon & Schuster, employees attended workshops with OpenAI representatives and the company held a contest offering $10,000 for the best AI use case, though it has since "rolled back some of the AI-pushing because employees kept complaining about it in town halls," one staffer said. Entry-level salaries at Big Five publishers and Scholastic averaged $47,583 in New York City in 2023.

The investigation finds that one of the most widespread applications is generating publicity copy—the text readers see on Amazon, Goodreads, and back covers. Literary agents began suspecting certain Big Five editors were using AI to write rejection letters after comparing messages on different books from the same editors and finding "the same classic signs of AI writing, like 'not this, but that' and the rule of three." Employees at all three publishers confirmed those suspicions to WIRED. Agents expressed fears that publishers might paste manuscript text into LLMs to generate rejection letters and publicity descriptions without using closed-loop models that prevent data sharing with AI companies' cloud servers. One HarperCollins source said multiple editors asked managers if they could input full manuscripts into open-loop LLMs and were told no by the legal team. Dan Sinykin, author of Big Fiction: How Conglomeration Changed the Publishing Industry and American Literature, said publishing executives "have more in common with the C-suites at other major American companies than they do with the editors and publicists and translators who work for them" and are "trying to see what they can get away with before there's a revolt."

Workers say they're turning to LLMs because executives slashed teams through layoffs and tripled workloads, with large teams reduced to five or six people handling double or triple the number of books. In recent weeks, Simon & Schuster employees learned the company was testing Skan AI—software that "gives leaders the evidence base they need to reduce costs, prioritize automation, and meet board-level transformation mandates"—at the urging of KKR, the investment firm that purchased the publisher in 2023. Staff began circulating an open letter opposing the surveillance and automation tool. CEO Greg Greeley responded with a memo saying no decision had been made, though AI-detection platform Pangram indicated at least some of the memo was AI-written. Most industry professionals interviewed want clear AI usage guidelines and disclosures at minimum. Emma Dries, an agent at Triangle House Literary who previously worked at HarperCollins and Penguin Random House, said that "short of these tools being banned, which would frankly be my preference," the industry needs more communication about when and how AI is deployed. Publishers may struggle to contain worker resistance as automation tools move from marketing tasks to core editorial functions, particularly if transparency remains absent. The path forward will likely depend less on technology capabilities than on whether leadership can reconcile efficiency gains with the human expertise that remains central to discovering and developing literary talent.