Text and data mining (TDM) for AI training is the starting point of a development process that can lead to powerful generative AI systems (GenAI). As these systems have the potential to replace human creativity and change human creative practice, the question is how copyright law could be used as a regulatory tool to support human free expression. With regard to the TDM process itself, the evolving ‘right to research’ under freedom of expression law has become an important factor in the policy debate. Moreover, the question arises how copyright can ensure fair remuneration for the use of human works in AI training – for both authors and rightsholders, including small repertoire owners. Finally, the quality and cultural diversity of TDM resources must be considered. The future generation of “GenAI natives” (not knowing a world without GenAI) may use GenAI routinely as a tool to obtain templates for their own creativity and free expression. If AI training resources are not sufficiently diverse, the risk arises that certain human expressions are no longer visible in GenAI output – and simply forgotten and lost when human acts of expression are based on GenAI templates. The session explores these different dimensions of the TDM and AI training debate.