Oxford University has granted OpenAI permission to use historical texts from the Bodleian Library for training its AI models. The collaboration, launched in March 2025, was initially presented as a digitisation project aimed at students and researchers.
Documents obtained through a freedom‑of‑information request show that part of the digitised collection has been added to OpenAI’s training datasets. The material comprises roughly 125,000 digitised images of historic theses and about 10,000 printed 16th‑century ballads containing lyrics and musical annotations. Further items slated for digitisation include 18th‑century Irish state papers, private letters of writer Marie Edgeworth, and Dorothy Hodgkin’s notebooks on penicillin research.
The arrangement has raised concerns among some members of the university community, who point to reputational risks of partnering with a controversial tech firm and the high energy consumption of large‑scale AI systems. Oxford maintains that the project is limited in scope, uses only works whose copyrights have expired, and that the university retains rights to the digital copies, which it plans to make publicly available online soon.
The deal mirrors a wider industry shift as AI companies turn to physical book and document collections after the proliferation of AI‑generated content diminishes the web’s value as a data source. OpenAI holds similar agreements with the Boston Public Library, MIT and the University of Michigan.
If expanded, the partnership could substantially increase the digitisation of the Bodleian’s estimated 23 million items, prompting a broader debate on the conditions under which libraries and universities should collaborate with AI firms to preserve and disseminate historical knowledge.