Artificial Intelligence and the Protection of Documentary Heritage

Training foundation models has turned libraries, archives and collections into strategic inputs. The article argues that the debate cannot be reduced to copyright licensing: digitisation for training operates on copies that are also cultural property, and it places on the custodian a duty of conservation that no copyright authorisation replaces. It distinguishes three legal planes (the work, the copy and the cultural value of the collection) and three purposes of digitisation (preservation, human access and training). It proposes a due diligence protocol based on prior conservation assessment, separation of purposes, traceability of copies and files, and reinvestment of consideration into conservation. The thesis does not presume that all digitisation is destructive or that all cooperation with developers is illegitimate; it holds that the decision must be documented, proportionate and compatible with the institution's public mission. English translation of the author's Spanish-language article; preprint, not peer reviewed.