Three Major Publishers Quietly Roll Out LLMs, and Staff Are Angry

Employees at three major publishing houses told WIRED that language models are already at work in their companies in four areas: advertising and PR, cover creation, back-cover copy and everyday correspondence. Some executives are also asking junior staff to become internal "champions" of the technology. There are no big announcements, and the rollout is happening quietly. Staff know about it and are unhappy.

We have only a retelling of the original piece and the words of anonymous sources, so the names of the companies and the scale of usage are outside the scope of this text. But the picture is telling: in an industry that sells other people's authorship, AI is entering through support tasks, not through the editor's desk.

Where the models already work

The choice of areas is easy to understand. Advertising materials, emails to journalists and bloggers, and blurbs on the back of the cover belong to a genre where volume matters more than authorial voice and deadlines are tight. This is an ideal target for LLMs: short format, templated structure, texts that are usually written in an hour and forgotten.

Covers are different. A cover is the face of the book and the workplace of an entire profession: illustrators and designers, many of them freelancers. When a publisher uses image generation, it directly replaces someone's commission. In 2024, Bloomsbury already faced a scandal over the cover of a book by Sarah J. Maas, in which readers found a generated element. The publisher acknowledged the problem and replaced the image. The lesson then was simple: readers notice this kind of thing quickly.

Why employees object

The resistance is growing for more reasons than fear of losing jobs. Publishing staff are, almost by definition, people who chose the profession for the sake of books and authors. Being asked to "promote" AI inside the team puts them in an awkward position: they have to persuade colleagues to use a technology that was trained on texts, possibly by the same authors they work with.

The backdrop to these disputes is heavy:

• In 2025, Anthropic agreed to settle a lawsuit by authors for roughly $1.5 billion over its use of pirated copies of books. That worked out to roughly $3,000 per work.
• Back in 2024, HarperCollins offered authors the chance to license their books for model training: about $5,000 per title, split equally between author and publisher.
• Authors' organizations in the US continue to sue model developers.

When a publisher tells authors "we protect your rights" while quietly plugging a chatbot into marketing, trust becomes the resource that is spent fastest.

What is known from the WIRED piece

• Three major publishers; the sources are anonymous.
• Four use cases: PR, covers, back-cover copy, email.
• Pressure on junior staff from some executives.
• Public statements on AI usage policy, as described, are either absent or vague.

The last point is probably the most important. Employees are angry less about the tool itself than about the fact that nobody showed them the rules of the game.

Practical takeaways

Authors and agents should add a direct question to their contracts: is AI used in creating the cover and promotional copy, and can they opt out? Illustrators and designers would do well to write into their contracts a ban on using their work to train models.

Publishers would be wise to publish a clear policy before a leak formulates one for them. Three points are enough: what the model does, what the human does, and where a label is applied. Executives who "push" AI through junior staff will have to explain how a person is accountable for the result if a neural network wrote the text. Without such an answer, the rollout will end not in savings but in a letter from an author, with the press copied in.