Microsoft inks deal with HarperCollins to train AI on books

How much is your life's work worth to a trillion-dollar tech company? If you ask HarperCollins, the opening bid is about twenty-five hundred bucks.

Earlier this week, 404 Media reported that publishing giant HarperCollins began asking authors if they’d license their nonfiction books to an unnamed tech company for AI training. Bloomberg quickly confirmed that the mystery partner is none other than Microsoft.

Under the deal, Microsoft plans to use selected nonfiction backlist titles to train an unannounced AI model. HarperCollins stressed that the program is strictly opt-in, meaning authors can turn it down without penalty.

A three-year lease on your brain

Author Daniel Kibblesmith blew the lid off the financials by sharing screenshots of the offer email. The terms: a flat $2,500 per book for a three-year license.

The contract reportedly includes a few guardrails. It promises author credit and puts strict limits on how much verbatim text the AI can spit out in its answers.

Kibblesmith called the proposal abominable and flatly rejected it. When asked what number would actually change his mind, he joked he’d do it for a billion dollars—or whatever amount would let him retire, since replacing human labor is basically the end goal of generative tech anyway.

A source familiar with the matter told Bloomberg that Microsoft does not intend to use the text to churn out AI-generated books. It simply wants high-quality factual prose to sharpen model performance. Microsoft itself declined to comment.

Content deals become the new normal

This isn't an isolated experiment. HarperCollins' parent company, News Corp, signed a massive deal with OpenAI back in May, giving ChatGPT access to archives from The Wall Street Journal and The Sunday Times.

While The New York Times is still fighting Microsoft and OpenAI in federal court over alleged copyright theft, other major publishers like Vox Media and The Atlantic have already decided that getting paid something beats fighting a losing war against web scrapers.

HarperCollins framed the initiative as modern publishing pragmatism. The company stated its job is to bring authors new monetization opportunities while protecting royalty streams and establishing model guardrails.

Why it matters

We think Microsoft's strategy here is clever on paper, but the execution exposes just how little big tech values individual creators.

For AI developers, high-grade web data is drying up fast. Edited nonfiction books are the holy grail for training factual reasoning and reducing hallucinations. By going through a major publisher with an explicit opt-in clause, Microsoft gets clean, legally bulletproof training data without risking another high-profile lawsuit.

Yet offering $2,500 for three years of access feels almost insulting to anyone who spent years researching and writing a book. For struggling backlist authors, that check might pay a month of rent, but it essentially asks them to subsidize the very systems designed to automate research and writing.

This opt-in blueprint is likely how the entire publishing industry will handle AI licensing going forward. It hands big publishers a steady new revenue cut while leaving individual writers to wrestle with an impossible dilemma: take the pocket change now, or hold out on principle while the rest of the world opts in.