Big Five Publishers Quietly Expand AI Use as Staff Push Back

Several of the largest U.S. book publishers have been quietly putting generative AI into everyday workflows even as they publicly warn about the technology’s risks, and that growing gap is now triggering resistance from employees, agents and authors. Staff at HarperCollins, Simon & Schuster and Hachette say AI is increasingly being used for publicity copy, agent emails, marketing materials and even cover design, while some workers are now organizing against broader surveillance-style AI tools.

The backlash matters because publishing depends on trust: authors share unpublished manuscripts, agents expect human judgment, and readers increasingly want to know when machine-generated text or art has been used. What is emerging, according to more than two dozen workers and agents interviewed about the industry, is a messy, largely undocumented rollout that has outpaced internal rules and public disclosure.

For months, major publishers have presented themselves as wary of generative AI, with some taking visible steps to challenge its use. But inside the Big Five, employees describe a different reality: executives are testing enterprise AI products, managers are encouraging “champions” to find use cases, and some teams are using large language models to speed up work after layoffs left them with thinner staffing and heavier workloads.

The result is an industry divide between the public stance and the internal practice. On one side are lawyers, editors and agents worried about copyright, confidentiality and the erosion of human craft. On the other are executives looking for efficiencies and employees trying to keep up with rising demands. The dispute now extends beyond content creation to workplace monitoring, with Simon & Schuster workers objecting to a trial of Skan AI, a software product that can map workflows and identify automation opportunities.

How publishers are actually using AI behind the scenes

AI use inside publishing is more widespread and mundane than many readers might expect. Workers across the Big Five say the most common uses involve administrative and promotional tasks rather than full-scale book writing.

At HarperCollins, Simon & Schuster and Hachette, employees say generative tools such as Claude and ChatGPT are being used to draft publicity emails, brainstorm marketing language, create back-cover copy and help generate assets for promotions. Some staffers also described AI-assisted cover art and marketing videos, particularly in divisions that have embraced the tools more aggressively than others.

That shift has changed how some publishing jobs feel in practice. Publicity copy is one of the most visible forms of publishing writing because it appears on Amazon, Goodreads, jackets, catalogues and media pitches. It shapes how books are introduced to readers and how they are sold to booksellers and reviewers. In that context, some publicists see AI-generated language as a threat to the credibility of the pitch itself.

“Using AI to generate publicity copy is short-sighted,” one Big Five publicist said, arguing that public-facing communications require real engagement with readers and media contacts.

Others in independent publishing believe the trend is likely to accelerate in large corporate houses. One publisher at a smaller press said the Big Five may be especially tempted to use AI in marketing and publicity because those functions are easy to standardize, even if the move conflicts with the idea that books are fundamentally human-made cultural products.

What kinds of tasks are being automated first?

The first tasks targeted are usually the ones that can be standardized, repeated and measured. In publishing, that includes press releases, email templates, jacket copy, internal summaries, social assets and early-stage brainstorming.

Workers said some editors have also experimented with using AI to draft rejection letters, a move that has angered literary agents. One agent said colleagues compared emails from the same editors and noticed recognizable machine-writing patterns, including repetitive phrasing and formulaic transitions. Employees at the publishers named in the reporting said those suspicions ring true.

There are also reports of foreign publishers and translation partners openly acknowledging AI use in some material shown at book fairs. Staff members say that, even where the use is disclosed, the shift raises questions about editorial oversight and whether publishers are trading judgment for speed.

Why employees are rebelling now

Employees say the pushback is not only about technology, but also about timing, transparency and pay. Many publishing teams have shrunk after layoffs, leaving the remaining staff with more books to manage and fewer people to do the work.

In that environment, some workers say they are being nudged toward AI because leadership wants more output with fewer people. One employee described the industry as running on “skeleton crews,” where teams that once handled a manageable list of titles are now expected to do double or triple the volume.

A worker at one of the large publishers said the real issue was not simply that staff were using shortcuts, but that management was effectively imposing those shortcuts in order to keep pace with cutbacks and revenue pressure.

That frustration is compounded by wages. Some publishing employees note that early-career salaries remain relatively modest compared with the cost of living in New York, where many publishing jobs are based. To some staffers, money spent on enterprise software feels like a poor substitute for paying people better or hiring more of them.

HarperCollins employees, according to those familiar with the internal meetings, were especially unsettled when leadership framed AI experimentation as a practical necessity. When some workers raised concerns about legality, ethics and the environmental cost of generative tools, they say those worries were brushed aside as part of the business cost of staying competitive.

What happened at HarperCollins?

HarperCollins appears to have been one of the more active test cases. Employees say senior leaders bought Claude licenses but struggled to define how the tools should be used, which led to the creation of internal “AI Champions” tasked with finding practical applications.

Those employees were asked to join brainstorming sessions and identify where AI might fit into company processes. Some staffers say the sessions quickly became a source of discomfort rather than excitement, especially after management added messaging to an internal AI page minimizing the environmental footprint of the tools.

Workers say the company has also acquired licenses for ChatGPT and Jasper, a platform that markets itself as capable of running marketing workflows from start to finish. One division was described as fully committed to using the software, including the creation of AI-generated marketing videos.

HarperCollins did not respond to requests for comment on the reported uses.

How are literary agents responding?

Literary agents are responding by tightening contract language and demanding more clarity from publishers. Their main fear is not just that an AI tool will produce clunky copy, but that an unpublished manuscript could be pasted into an open system and exposed to a third-party model provider.

That concern has led some agencies to add contract clauses prohibiting material from being run through an LLM. Agents say the language is meant to protect authors from outside tech firms, but it also limits what a publisher can do with a manuscript once it is under contract.

One agent explained that the key distinction is whether a model is closed-loop or open-loop. In a closed-loop setup, the tool runs on an internal system and does not send data back to a cloud service. In an open-loop system, the writing may be processed by a vendor’s servers, raising questions about confidentiality and ownership.

According to one source at HarperCollins, editors have asked whether they could paste full manuscripts into open models and were told no by the company’s legal team. That detail suggests publishing lawyers are already trying to draw boundaries, even if many staffers want those boundaries spelled out more clearly.

What is the Skan AI controversy at Simon & Schuster?

The Skan AI episode is about more than software; it is about the fear that monitoring tools can be used to measure people out of their jobs. Simon & Schuster staffers say they heard the company was testing Skan AI at the urging of owner KKR, which bought the publisher in 2023.

Skan AI says on its own website that it helps leaders identify ways to reduce costs, automate processes and meet board-level transformation goals. That language alarmed employees, who interpreted the software as a surveillance and automation product that could map out their work and then replace parts of it.

In response, workers circulated an open letter to CEO Greg Greeley calling for a commitment not to introduce Skan AI or similar tools into the company’s workflow. They also objected to the broader implications of software that can expose patterns in day-to-day work.

Greeley later told staff that the company had only been exploring the product and had not made a final decision. He said he had not seen the materials until concerns were raised to him. That memo, however, was itself scrutinized by employees who used AI-detection tools and believed portions of it may have been machine-written.

Why do monitoring tools matter so much in publishing?

Monitoring tools matter because publishing relies on intellectual labor that is difficult to measure in simple productivity metrics. Editorial judgment, relationship management, and long-term acquisitions work do not fit neatly into automation dashboards.

When staffers hear that a company is testing software designed to expose bottlenecks, they worry the next step will be restructuring by algorithm. That is especially sensitive in a business already marked by layoffs and consolidation, where employees feel they are constantly being asked to do more with less.

For many workers, the issue is not only privacy but also the culture of trust. If a publisher begins using tools that track every process, staffers say, it can undermine the sense that editorial work is being valued for judgment rather than reducible to metrics.

How do publishers justify the shift?

Publishers generally frame AI as an operational tool, not a replacement for creative judgment. That distinction is important because it lets companies claim they are not using AI to write books, even if they are using it to produce supporting materials around books.

Hachette says it supports AI for operational purposes when it helps books reach more readers, but not for creative uses such as communicating with authors and partners. Simon & Schuster says employees have access to a limited set of vetted enterprise tools, but use is not required. The company says tools must be approved and that the goal is to reduce busywork so employees can focus on deeper, more creative tasks.

Those statements reflect a common corporate defense: AI is just a productivity aid, and staff will benefit from having more time for higher-value work. Critics inside the industry counter that the tools are not being introduced to enrich employees’ creative time, but to lower labor costs and increase output.

That tension is visible in how executives talk about the future. HarperCollins chief executive Brian Murray has previously said AI offers more opportunities than risks, especially in translation and audiobooks. His comments fit a broader industry argument that machine-assisted translation could help publishers scale international publishing more quickly.

Not everyone agrees. Some translators and editors worry that the push for speed will undercut quality and erase the human expertise that gives books their cultural and linguistic texture. Others note that even when AI is used in translation, the sector has historically relied on invisible human labor, and more disclosure should be the minimum standard.

Why translation and cover art are becoming test cases

Translation and cover art are natural pressure points because both involve a mix of technical skill and creative judgment, and both can be partially assisted by software. In the translation market, staffers say some manuscripts have already been shown with AI-use disclosures at major European rights fairs.

For publishers, the appeal is obvious: faster turnaround, lower costs and potentially more books in multiple languages. But the risk is equally obvious: the more a publisher relies on machine output, the more readers and rights holders may question the integrity of the final product.

Cover art is another area where AI can quietly enter the workflow. Employees say some designs and promotional visuals have already been touched by machine tools, including at foreign presses linked to the same corporate networks. Because cover art is often the first thing a reader sees, any perceived shortcut can affect how the entire title is received.

The controversy also raises a branding issue. Small and independent presses often market themselves on curation, human taste and editorial care. If Big Five firms standardize more of their visual and promotional output, smaller houses may be able to distinguish themselves by emphasizing handcrafted work and closer author relationships.

What are workers and experts asking for?

For now, the most common demand is not a total ban but a clear, public set of rules. Editors, publicists and agents want publishers to define what can be put into AI systems, what must never be uploaded, and when disclosure is required.

That includes obvious questions:

• Can an unpublished manuscript ever be entered into a third-party model?
• Can AI be used to draft publicity copy without disclosure?
• Should authors be told when AI helps create a cover, pitch or marketing campaign?
• What kind of data protections apply to internal trials and workflow tools?

Some workers say they would prefer an outright prohibition on generative AI in publishing. Others are willing to accept limited use if the rules are strict, the tools are air-gapped or closed-loop, and authors are informed when their work is touched by machine systems.

Emma Dries, an agent who previously worked in editorial, said publishers need to communicate more clearly, not less. Her view reflects a broader sentiment across the industry: if AI use is going to happen, hiding it is the worst possible strategy.

Why the industry is headed for a bigger fight

The clash inside publishing is likely to grow because it is part of a wider corporate pattern. Large companies often test new tools quietly, then expand them incrementally before explaining them publicly. By the time workers notice, the technology is already embedded in daily operations.

That is particularly risky in a business where the product is information and the raw material is often confidential. If AI tools are trained on or exposed to unpublished manuscripts, the consequences are not abstract. They go to the heart of author trust, contract enforcement and the value of literary labor.

Observers of the industry say the current moment also reflects a deeper philosophical divide. Corporate leadership often treats AI as a way to optimize workflows, while editors, translators and publicists see their roles as built around discretion, taste and human relationship-building.

As one publishing scholar put it, the people making decisions at the top of conglomerated publishing houses increasingly resemble executives at any other major American company: they are looking for efficiency, return on investment and leverage over labor. The fear among staff is that the publishing sector is trying to see how much it can automate before readers and workers push back.

One industry observer argued that the big houses are not merely adopting new software; they are testing the limits of what the sector will tolerate before there is a full-scale backlash.

What comes next for Big Five publishers?

The next phase will likely depend on whether companies choose transparency or continued ambiguity. If publishers publish clear AI policies, restrict manuscript uploads, and disclose machine involvement in marketing or design, they may be able to slow the backlash. If they continue to expand use quietly, more staff may follow the Simon & Schuster example and organize publicly.

What is already clear is that the book industry is no longer debating AI in the abstract. It is dealing with real products, real trials and real internal conflicts. The issue now is not whether generative AI will enter publishing, but who gets to decide how, where and under what rules.

For authors, the immediate question is whether their manuscripts are safe and their names are being represented accurately. For workers, it is whether the push for efficiency will further erode staff size, wages and editorial control. And for readers, it is whether the books and book promotions they see will remain meaningfully human-made, or become just another corporate experiment in automation.

In an industry built around voices, the most important question may be whether the people making books will still be allowed to speak for themselves.