HarperCollins

HarperCollins, the nearly two-century-old publisher owned by Rupert Murdoch’s News Corp, broke ranks with the rest of the Big Five in November 2024: rather than suing or stonewalling AI companies, it sold to one. The three-year deal — confirmed after 404 Media surfaced it, with Bloomberg identifying Microsoft as the buyer — licenses select nonfiction backlist titles for AI model training.

The first price tag on a book

The terms set the first public market price for a book as training data: $5,000 per title, split evenly, with $2,500 going to the author regardless of advance status. Participation is opt-in — authors who decline stay out of the dataset — and the contract caps model outputs at 200 consecutive words or five percent of a book’s text, alongside a pledge not to scrape pirate libraries. The The Authors Guild cautiously credited the opt-in structure and 50-50 split while questioning whether the price was adequate and warning authors to read the fine print; agents complained the flat fee treated a Pulitzer winner and a diet book identically.

Author reaction was loud. Novelist Daniel Kibblesmith publicly mocked the offer he received, and critics argued the deal legitimized training on books just as authors’ lawsuits against AI firms were gaining ground. Others saw the pragmatic logic: models were being trained on pirated books anyway, and HarperCollins had at least converted that into consent and cash — the opposite bet from Penguin Random House, which wrote a no-training clause into every copyright page the same season.

CEO Brian Murray has since framed AI as both threat and opportunity for publishing, and the deal remains the template every other trade publisher is measured against — the first real experiment in what a licensed, author-consenting training market looks like.