University of Oxford Partners with OpenAI for Library Digitization and Model Training

The University of Oxford has entered a partnership with OpenAI to digitize historical texts from the Bodleian Library. Internal documents confirm that the digitized material is being utilized to populate OpenAI’s machine-learning training sets.

University governance committees have formally recorded concerns regarding the reputational risks associated with this collaboration. Staff have also raised questions regarding the environmental impact of the energy-intensive computational processes required for AI model training.

University officials maintain that the project is limited to out-of-copyright materials and that the institution retains ownership of the digital scans. The university plans to release these materials for public access in the coming months, asserting that the partnership facilitates broader research availability.

This agreement is part of a broader initiative involving several international research institutions. The project highlights the increasing reliance of AI developers on physical, historical archives to secure high-quality, non-synthetic data for model development.