This presentation explores how European and German copyright law apply when AI systems are trained on large datasets that may include protected works. While such uses can amount to copyright infringement, the German text and data mining exceptions in §§ 44b and 60d UrhG offer important — but contested — legal bases for automated data processing. It outlines, when AI training may qualify as (scientific) text and data mining, highlights current debates in scholarship and case law, and discusses what researchers should know to avoid copyright risks. It was held in context of the Interdisciplinary Colloquium on Digitalisation of Research of the Leibniz Science Campus "DiTraRe" (=Digital Transformation of Research)