Microsoft has formed a licensing arrangement with HarperCollins to bolster its AI models by training them on nonfiction literature. As per reports from *Bloomberg*, this agreement permits Microsoft’s large language models (LLMs) to utilize premium nonfiction material, although the company is said to have no intentions of producing AI-generated publications. The primary aim is to enhance the intelligence and precision of its models. Authors associated with HarperCollins will have the choice to join the initiative or decline participation, as indicated in the report.

The initial announcement of the agreement was made by [404 Media](https://www.404media.co/harpercollins-ai-deal/), which first characterized it as a licensing pact with an undisclosed AI organization. Author Daniel Kibblesmith posted images of an email, presumably from his agent, on Bluesky, that detailed the deal. The correspondence acknowledged the ongoing debate regarding the utilization of copyrighted works for AI training, particularly when organizations neglect to remunerate or acknowledge the original creators. The message also raised alarms that AI models might ultimately supplant human artists.

Kibblesmith voiced his discontent with the announcement on Bluesky, labeling the agreement as “abominable.”

According to the email images, HarperCollins is proposing a fixed payment of $2,500 per title for a three-year licensing deal to authors. HarperCollins confirmed the existence of a contract with an unnamed AI firm, informing *Bloomberg* that the agreement is of “limited scope” and incorporates “clear guardrails around model output” to uphold authors’ rights while also offering them fresh opportunities.

Simultaneously, various reports from sources like *Bloomberg* and *The Information* have highlighted that AI companies, such as Google, OpenAI, and Anthropic, are encountering decreasing returns in crafting more sophisticated models due to insufficient high-quality data for training. While Microsoft was not specifically mentioned in these reports, its Copilot model, which relies on OpenAI’s generative AI technology, may gain from this new collaboration with HarperCollins as a means of tackling the issue of limited data.