Canada

‘Swamped by bots’: University libraries raise alarms over data-scraping AI systems

Published: 

Peter Lewis from Ontario Tech University discusses university websites facing increasing pressure from AI bots scraping huge swaths of data.

University library websites are being overwhelmed by bots that are scraping data to train generative AI models.

“AI models are trained on huge amounts of data,” Peter Lewis, an associate professor in computer science at Ontario Tech University, told CTV’s Your Morning on Tuesday.

“Libraries are raising the alarm bells that they’re being swamped by bots that are collecting this kind of data through automated searches and queries at a scale that’s actually taking out services in some cases and putting public access at risk.”

Lewis, who is the Canada Research Chair in Trustworthy Artificial Intelligence, says there are two main reasons that companies are trying to scrape data from academic resources like library websites in order to train AI models such as OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini.

“In general, when you’re training general purpose models, you want to have a broad set of high-quality content,” Lewis explained. “The second one is we’re increasingly looking to train models on what might be sort of niche or really specific kinds of topics.”

In just one example of this, the University of North Carolina at Chapel Hill reported a staggering 11,329 searches for Finnish music during just one day in 2025.

Lewis says these kinds of searches can drive up costs for academic institutions and make it harder for legitimate researchers to access the resources they need. To that end, the University of Victoria recently announced that it would block autonomous AI agents from accessing university information systems.

“They’re often, for example, executing quite complex queries, which have a much greater toll on the back end of the library system,” Lewis said of AI bot searches.

“What that means then is that the servers are overloaded and it can behave a little bit like what we might call a ‘denial of service attack,’ where certainly any human users trying to access that resource receives a poor quality of service or the site just goes down and requires somebody to bring it back up again.”

While some companies have agreed to honour requests from institutions that don’t want their data to be scraped to train AI systems, Lewis says such companies could also simply pay institutions for the data they’re seeking. That could work as individual charges for searches or even broader agreements between academic institutions and AI companies.

“Perhaps more promising is the idea of actually trying to agree on platform charging models, so that a particular resource or public institution that has been created to hold this kind of information has a platform agreement with a company that wants to access that, and then they can agree on things like the quantity and the load and things and plan for that as well,” Lewis said.

“I think that might be what I would prefer as a long-term solution.”