Oxford Lets OpenAI Train AI on Bodleian Library Books

Oxford University in the UK has allowed OpenAI to use historical books from its Bodleian Library to train AI models. An internal document shows that digitized Bodleian collections processed by OpenAI have been used to build training sets. Oxford announced the partnership in March 2025, but the announcement did not mention the materials would be used for model training.

Meeting minutes obtained under Freedom of Information laws show that faculty members, including Bodleian management committee members, raised concerns. They worried about reputational risks and noted that the high-energy technology could affect the university's environmental commitments. Oxford denied hiding the machine learning purpose, saying digitization was the core goal and faculty always acknowledged the project would provide training data.

As of June 2025, the Bodleian has provided OpenAI with 125,000 scanned pages of historical dissertations from 19th and 20th century European and American universities, plus a rare 16th-century broadside ballad collection. Oxford is the only UK member of OpenAI's NextGenAI project. The university says the digitized materials are out of copyright, OpenAI's access is non-exclusive, and the library retains the right to publish the resources independently.