Oxford let OpenAI train its models on Bodleian Library texts, documents reveal

[Highland Canopy]

The Guardian has reported that Oxford University let OpenAI use texts from the Bodleian Library to train its AI models. When the partnership was announced in March 2025, it was framed as a project to digitise material and make scholarship more accessible. Internal documents obtained through freedom of information requests show that training ChatGPT was part of the arrangement too. By June 2025, around 125,000 scans of old theses had reportedly been sent to OpenAI

The material involved is historical and out of copyright, which is Oxford's main defence. The university says the AI training side was not hidden and that staff knew the texts would support model development alongside the scanning work. The Bodleian keeps the rights to the scans and intends to publish them online for anyone to use. Oxford is the only UK member of OpenAI's NextGenAI consortium, which also includes MIT, Caltech, the University of Michigan and Boston Public Library

According to the documents, some staff raised concerns about reputational damage and the energy use of AI. There were also worries about outdated or offensive language in old texts ending up in training data. Those are fair points to raise. OpenAI's line is that with over a billion people using the technology, it is important for it to reflect different cultures, histories and perspectives. The consortium as a whole received $50 million in research grants and computing funding

I have mixed feelings about this. On one hand, the texts are out of copyright, the public gets free digitised versions and nobody's current work is being taken without permission. On the other, it feels a bit off that the training part was not front and centre when the deal was first announced. An institution like the Bodleian trades on trust, and people like to know what their libraries are doing

Is this a sensible deal that benefits everyone, or should universities be more careful about handing their collections to AI firms? Would it have been different if they had been upfront from day one? Keen to hear from anyone who works in libraries

Oxford lets OpenAI train its AI models on Bodleian Librarywww.theguardian.com

[Rooney]

Scanning rare material and putting it online for free is a public good, whatever else is going on. A lot of these texts would never have been digitised otherwise, because libraries do not have the money. If AI money pays for access, maybe that is not the worst trade

The real test is whether the scans do go online. If they do, most people will forgive the rest. If not, the criticism is deserved

[IonStorm99]

The problem is not the deal itself, it is the way it was presented. Calling it a digitisation project and leaving out the training part is a PR choice

When it comes out through FOI requests two years later, it looks like something was hidden. Being open at the start would have avoided all of this

[QuietObserver13]

Staff worrying about outdated language is a fair point. Nineteenth century writing on other cultures can be pretty grim by modern standards. Feed enough of that into a model and it could reflect those views in odd places. Hopefully OpenAI filters for it. I would not bet on it though

[Dylan70]

Out of copyright is out of copyright. Anyone could have scanned those theses and trained a model on them. At least this way the public gets free access to the digitised versions as well

[IronWolf]

Fifty million dollars split across fifteen institutions is not a lot for what these companies are getting. Libraries have spent centuries collecting and preserving this material. The tech firms get it for the cost of some scanning and a few grants

That said, grants are better than nothing. Plenty of libraries get no help at all

[ElPresidente]

Old theses are a goldmine of carefully written, well sourced material. That is exactly the kind of text AI companies want more of, because so much of the web is low quality. I can see why OpenAI was keen. Oxford should have negotiated a much better price for it