Publishers Are Feeding Your Manuscript to ChatGPT. The Authors Guild Is Not Amused.

Here is a fact that may unsettle your Tuesday: somewhere in a publishing house, an editor has almost certainly pasted a portion of an unpublished manuscript into ChatGPT. They may have called it "proofreading." They may have called it "developmental feedback." Whatever they called it, they did not ask the author.

The Authors Guild, which has spent the past year navigating the wreckage of the Anthropic copyright settlement, issued a statement this week addressing something arguably more intimate than training data: the quiet, daily habit of editorial AI use. The trigger was a Bookseller report documenting exactly this behavior -- editors uploading manuscripts to consumer-facing chatbots without disclosure or permission. The Guild's response was unambiguous: doing so "may constitute a violation of the author's copyright or right of privacy."

Umair Kazi, the Guild's director of policy and advocacy, made the distinction that matters: publishers have started inserting boilerplate AI clauses into contracts -- language that permits AI use "in their regular practice" -- without specifying what that actually means. Spelling and grammar checking is one thing. Feeding a debut novelist's first chapter to a model that may regurgitate fragments of it three months later is quite another.

The Guild is now pushing for sandboxed models with guardrails and written author consent for any editorial AI use. Whether that standard becomes industry norm or dissolves quietly into the stack of worthy intentions is another question. Hachette, to its credit, recently cancelled publication of a book called Shy Girl after discovering AI-generated content -- which suggests at least some houses are paying attention.

There is something depressingly familiar about this moment. Every new technology that touches the book industry follows the same arc: enthusiasm, overreach, quiet damage, belated regulation. The difference now is the intimacy of the violation. A manuscript is not a dataset. It is, quite literally, someone's unpublished thought. The fact that we need a formal statement to make that point suggests the industry has further to fall before it finds the bottom.

Authors: read your contracts. All of them. Especially the parts that seem boring.