Princeton University Press Signs AI Licensing Deal With Cashmere, Bars Catalog From LLM Training

Princeton University Press has signed a partnership with Cashmere, an independent AI content-licensing platform, to make its catalog available to what the companies describe as "copyright-compliant, non-LLM" AI platforms. The press announced the deal on August 3.

The agreement follows Cashmere's existing licensing relationships with Rockefeller University Press and Springer Publishing, extending a small but growing model for how publishers route content to AI systems without feeding it into general-purpose model training.

Cashmere's pitch to publishers rests on a distinction the AI industry rarely draws in public: content used for retrieval-augmented generation, where a system fetches and cites a passage on demand, is licensed and metered separately from content absorbed into a model's training weights. The platform's terms rule out the latter outright.

Under the agreement, Princeton University Press keeps control over which titles are licensed and on what terms. The underlying book contracts and copyrights are unaffected — the press is licensing access to content it already controls, not signing rights away.

Book contracts and their underlying copyrights are fully respected under the arrangement.

— Princeton University Press / Cashmere announcement

For working authors, the deal is a data point in the still-unsettled question of who gets paid when an AI system draws on their words. It matters most directly to academic authors whose contracts assign subsidiary-rights and licensing decisions to the press rather than to them individually. Trade and indie authors whose publishers haven't struck a comparable deal remain outside any licensing arrangement — paid or not — when their work turns up in a retrieval system's answers.