The largest book publishers are publicly policing artificial intelligence in authors’ work while adopting many of the same tools inside their own operations. WIRED reported on October 9 that HarperCollins, Simon & Schuster and Hachette have introduced generative AI into parts of publishing without broad public disclosure or author consent, creating a widening conflict between executive efficiency goals and the people responsible for editing, marketing and selling books.

The report is based on interviews with more than two dozen publishing workers who spoke anonymously because they feared professional consequences. They described large language models being used for publicity and back-cover copy, emails to literary agents, marketing materials and some cover art. Staffers also told WIRED that well-known authors’ projects have been canceled over undisclosed AI use without those incidents becoming public, suggesting the industry’s visible controversies represent only part of the dispute.

An AI token dominates a publishing meeting table surrounded by manuscripts.
Employees assigned to champion AI tools are raising legal, ethical, environmental, and labor concerns about their rollout.

At HarperCollins, senior leaders bought Claude licenses and then recruited junior and senior employees to act as “AI Champions,” according to current staff cited by WIRED. The employees were expected to identify potential applications in recurring brainstorming meetings. When participants raised legal, environmental and ethical objections at one session, a meeting leader characterized those concerns as a business cost, one attendee said. Workers viewed the investment in tools against a backdrop of longstanding complaints about low pay.

Publicity copy has emerged as a common application across the three publishers. That material appears on retail pages and book jackets and is used to pitch titles to booksellers, journalists and reviewers. One Big Five publicist told WIRED that automating those messages risks breaking the good-faith relationship between publishers and their audiences. Literary agents also said they had noticed formulaic rejection emails and suspected some editors were using AI to compose them.

The companies offered different accounts of their policies. Hachette told WIRED it supports AI for operational purposes that help books reach readers but rejects creative uses, including communications with authors and partners. Simon & Schuster said employees can access a limited set of approved enterprise tools, that their use is optional and that the aim is to create more time for focused and creative work. HarperCollins did not respond to the publication’s questions.

A publishing desk is overwhelmed by books beside a rapidly moving automation conveyor.
Workers say layoffs and larger workloads are making automated shortcuts tempting even for employees who distrust the technology.

Confidentiality is a central unresolved issue. Agents told WIRED they have added contract clauses prohibiting manuscripts from being run through large language models, rules that apply to publishers as well as technology companies. Their concern is that unpublished writing could be sent to a cloud-based system while generating descriptions or responses. A HarperCollins source said editors who asked about pasting full manuscripts into an open model were told by the legal team not to do so, but workers and agents still want clearer guidance and disclosure.

Employees also linked AI adoption to reduced staffing and rising workloads. Some told WIRED that layoffs had left small teams handling two or three times as many books, making automated shortcuts tempting even for reluctant users. The same tension surfaced at Simon & Schuster after workers learned about an experimental trial of Skan AI, a product that analyzes workflows for automation opportunities. Employees began gathering signatures for a letter opposing the product and similar monitoring systems.

Simon & Schuster CEO Greg Greeley later told employees that the company had explored Skan AI’s Blueprint product but had made no decision to adopt a productivity-monitoring or workflow-analysis tool, according to a memo described by WIRED. The episode captures the industry’s larger problem: policies are arriving after experimentation, while authors, agents and employees do not know consistently where AI is being used. The immediate demand from many workers is not a technological verdict but basic transparency—clear boundaries, informed consent and disclosure before automation becomes part of the publishing process.