Oxford Partners with OpenAI for AI Model Training at Bodleian Library

The University of Oxford has entered a significant agreement with OpenAI, the company behind ChatGPT, allowing its AI models to be trained using historical texts from the famed Bodleian Library. This collaboration aims to enrich AI with academic resources as tech companies increasingly seek valuable data from universities.

Documents indicate that the texts digitized by OpenAI will be incorporated into their training dataset. This partnership, announced in March 2025, is designed to use OpenAI’s technology to digitize unique materials from the Bodleian, enhancing accessibility for students and researchers alike.

While the initial announcement did not clarify that these materials would feed into OpenAI’s model training, the purpose is to enable the AI to analyze vast amounts of data to generate coherent sentences and perform other cognitive functions.

An OpenAI spokesperson expressed pride in ensuring that “the AI models of today preserve the world’s historical knowledge for the future,” highlighting the importance of diverse cultural perspectives in this technology accessed by over a billion users daily.

Internal documents revealed concerns from Oxford staff, particularly members of the Bodleian governance committee, regarding the potential reputational risks associated with partnering with OpenAI. They raised questions concerning environmental implications tied to the energy demands of AI technologies.

In light of this, booksellers have reported an increase in orders for rare and obscure titles, such as agricultural guides from 18th-century Africa or biographies of 1950s automobile racers. Bookshop owners suspect these titles, often not available online, are sought after for their unique data that can contribute to training future AI models.

The rising prevalence of AI-generated content online has rendered many scraped websites less effective for training purposes. Consequently, AI developers are now exploring physical and historical book collections as vital data sources.

OpenAI has pursued similar partnerships with various US research institutions, including the Boston Public Library, Caltech, MIT, and the University of Michigan under a project dubbed NextGenAI. Oxford University is the first and only UK institution involved in this initiative.

By mid-2025, OpenAI had received 125,000 images from the Bodleian’s collection, including 19th and 20th-century European and American PhD theses. The project also encompasses the digitization of 16th-century ballads, personal letters of notable authors like Marie Edgeworth, and scientific notes from renowned chemist Dorothy Hodgkin.

The contract between OpenAI and Oxford raises the prospect of expanding the digitization of the Bodleian’s extensive collection of 23 million items. Discussions include potentially creating an “Ask the Bod” chatbot for user engagement.

A representative from the University clarified that the current scope of digitization is “modest” and limited to materials not under copyright. The Bodleian retains scanning rights and plans to publicly release these digitized resources shortly.

They dismissed allegations that the machine-learning aspect was concealed from the public or students, emphasizing that digitization was the primary goal, while also contributing to OpenAI’s training dataset.

Importantly, the integrity of the Bodleian’s collections remains safeguarded under this agreement, unlike cases elsewhere where books have been destroyed after being digitized. Competing firm Anthropic has reportedly spent considerable resources acquiring books, which are then dismantled for scanning purposes, although they claim not to target rare books for destruction.

Furthermore, a tech news outlet observed practices where secondhand books were tracked to areas where they were dismantled and scanned.

The Oxford spokesperson reiterated that the content digitized in collaboration with OpenAI is modest, out-of-copyright, and not exclusive to OpenAI’s usage. The library maintains rights to publish and provide access to the digitized materials, furthering their mission to broaden public access.