Large language models (LLMs) have quickly become integral to historical research. Their ability to process, annotate, and generate texts is reshaping scholarly workflows. However, historians are uniquely positioned to ask a more profound question: who owns the tools that influence our understanding of history? Most of the most powerful LLMs today are created by private companies. Although these companies invest heavily, their objectives—centered on profit, platform expansion, or intellectual property control—often conflict with the core values of historical scholarship: transparency, reproducibility, accessibility, and cultural diversity. This situation raises serious concerns about a) opacity: we frequently lack clarity regarding training data and inherent biases, b) instability: terms of access and functionalities can change without warning, and c) inequity: many researchers, especially those in under-resourced environments, are excluded. It is imperative to develop public, open-access LLMs for the humanities, trained on carefully curated, multilingual, historically grounded corpora drawn from libraries, museums, and archives. These models need to be transparent, accountable to academic communities, and backed by public funding.

Creating such infrastructure is challenging but essential. Just as national archives or educational curricula would not be outsourced to private companies, neither should our most powerful interpretive technologies. The humanities bear both a responsibility and an opportunity to develop culturally informed, academically rigorous artificial intelligence. We must not only use LLMs responsibly but also take ownership of them responsibly. The integrity of scholarship and the future of public knowledge may depend on it. Prof Dr Matteo Valleriani Max Planck Institute for the History of Science, Berlin, Germany If you have an opinion on any item you’ve read in the Guardian today, please email your letter for consideration in our letters section.