Publishers Police AI on Authors. Inside, They Run It Undisclosed.
At least three of the Big Five US book publishers, HarperCollins, Simon & Schuster and Hachette, have been incorporating AI tools into their publishing processes without public disclosure or author consent. That is the central finding of a Wired investigation by Adam Morgan, built on interviews with more than two dozen workers who all spoke on condition of anonymity.
The same houses enforce AI rules on the people who supply them with books. One editor told Wired that “every single imprint has a story of someone very esteemed who had a book canceled because of blatant AI use.” A literary agent told Wired: “We’ve added AI clauses to our contracts saying ‘this material can’t be run through an LLM’.”
So the rule points outward. Authors who use AI can lose a book. An agent writes clauses to keep manuscripts away from models. Meanwhile, according to the workers Wired spoke with, AI runs inside the building without public disclosure or author consent.
We read this as a governance failure with a specific shape: an asymmetric policy, enforced on counterparties and waived internally.
What the record actually shows
Most of the core claims rest on anonymous sourcing. That matters, and it should temper how anyone repeats the story. The on-record policy statements are narrower and come from two companies.
Hachette’s policy, as quoted by Wired, says it supports “the use of AI tools for operational purposes” and does “not support creative uses of AI, including communicating with our authors and partners.” Simon & Schuster prohibits any AI tool “that has not been vetted and approved.” HarperCollins did not respond to requests for comment.
Read those two policies closely. Both govern which tools staff may use. Neither, in the text Wired quotes, tells anyone what the author will be told. Hachette draws its line around creative use and around communication with authors. The quoted text says nothing about informing authors when operational AI touches their work. Simon & Schuster’s line is an approval gate. An approved tool can still be used on a manuscript without the author hearing about it.
A HarperCollins staffer described the result plainly. There has been no direction “about what to say to authors or agents about any of this, because I think they know it would start a shitstorm.”
That sentence is the whole problem in one line. The staffer believes leadership expects friction from disclosure, and the reported outcome is silence where a policy should be.
Some of the controls already exist
The story contains evidence that these organizations can draw lines when they want to.
At HarperCollins, editors asked to paste a full manuscript into an open-loop LLM “were told not to by the company’s legal team.” That is a real control. Legal said no.
Simon & Schuster’s approval requirement is another real control. An allowlist of vetted tools is the right primitive for any company that wants AI use to be inspectable.
And the agent’s contract clause is a third control, written from the other side of the table.
So the raw material for governance is present. What is missing is the rule that connects these controls to the counterparty. Legal blocks one workflow. An approval gate vets tools. An agent writes prohibitions. None of those mechanisms produces a sentence an author can read that says: here is where AI touched your book.
Silence has costs inside too
Staff are pushing back on the adoption itself, which in our reading means the quiet approach is not cheap internally either.
In May 2024, Simon & Schuster ran an AI use-case contest with “a $10,000 prize for the winner and $5,000 each for two runners-up.” Wired reports the company later rolled back “some of the AI-pushing because employees kept complaining about it in town halls.” Staff also circulated an open letter seeking “a commitment to not bring Skan AI or other similar software into S&S workflows.” A CEO memo stated that “no decision has been made to use Blueprint or any other productivity-monitoring or workflow-analysis tool.”
Our reading is that there is pressure behind the adoption. One worker told Wired that “large teams have been reduced to just five or six people, give or take, and we’re doing double or triple the number of books.” Teams stretched that thin are likely to reach for tools whether or not a policy covers them. When the policy covers tool approval and stays silent on disclosure, the tools spread and the silence spreads with them.
Emma Dries of Triangle House Literary put the counterparty demand in one sentence: “We need publishers to be clear about what is being used and when.”
That request is modest. A record of tools and moments would satisfy it.
One rule, both directions
We have written about where the public and an industry framework each draw the AI disclosure line, about brand voice and disclosure on LinkedIn, and about HR becoming the function that authorizes AI. Publishing adds a variable those pieces did not isolate: the direction a rule points.
A disclosure rule that applies only to counterparties is a liability waiting for a reporter. Every contract clause a company imposes on suppliers becomes evidence of what it knew, once its own undisclosed use surfaces. The publisher that cancels a book for AI use has already conceded that AI use is material to the work. It cannot then argue that its own AI use on the same work is too minor to mention.
The fix is symmetry. Whatever disclosure you demand from a counterparty, you owe in the opposite direction, at the same level of detail.
Three controls worth adopting
The story hands any organization three controls. Each needs one adjustment to stop pointing one way.
The no-LLM clause. An agent wrote it. Companies can adopt it as a commitment they make, in addition to a demand they place on others. If you tell suppliers their material cannot be run through a model, publish the conditions under which you will run their material through one, and stick to them.
The vetted-tool allowlist. Simon & Schuster already requires approval. Extend the allowlist record with one field: which counterparty materials each approved tool may process. That turns an internal approval record into the source for external disclosure.
A disclosure rule written for the counterparty. Hachette’s policy speaks about creative use, including communicating with authors and partners. Write the version addressed to the author, the client or the supplier, in their language. It answers what Dries asked: what is being used, and when.
Do this now
Pull every AI clause your company imposes on vendors, contractors, freelancers or clients. Put them side by side with your internal AI use policy. For each clause, ask whether your own use of AI on that counterparty’s material would pass the same test you impose on them.
Where it would fail, you have found an asymmetry. Close it by writing the counterparty-facing disclosure first, then aligning the allowlist so every approved tool maps to what you will tell the people whose work it touches. If the honest answer to “what would we tell them?” is the one the HarperCollins staffer described, that silence avoids a fight, then that answer is the risk you are carrying.
This analysis synthesizes Book Publishers Are Quietly Using More AI. Staff Are Revolting (WIRED, October 2026) and As They Supposedly Battle AI Slop, Major Publishers Reportedly Cover Their Books in It (Gizmodo, October 2026).
Victorino Group helps companies write AI disclosure rules that hold up in both directions, for staff and for counterparties. Let’s talk.
All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →
If this resonates, let's talk
We help companies implement AI without losing control.
Schedule a Conversation