The Three-Step Test in International Copyright – A Global Framework for Generative AI Training

The training of generative artificial intelligence (genAI) models involves the processing of large datasets, which often include copyrighted materials, thereby raising questions of copyright infringement. These questions tend to center on whether such use constitutes fair use or falls under text and data mining (TDM) exceptions. In the US, the fair use doctrine-assessed through four flexible factors, including market effects-remains unresolved with regard to genAI training.

Meanwhile, in the EU, there is an ongoing debate about the applicability of the DSM Directive's TDM exceptions to genAI. What has remained largely unexplored in these discussions, however, is the fact that both systems must operate within binding international treaty constraints, particularly the three-step test requiring that exceptions do not unfairly conflict with copyright holders' economic rights. Any proposed reforms to copyright policy must therefore fit within these constraints, balancing innovation with rights protection.

This paper critiques broad copyright exceptions for genAI training and proposes conditions such as transparency, audits, and compensation to harmonize with international norms, while recognizing the nuanced differences between US and EU approaches and suggesting targeted legislative actions in the EU to align with the three-step test.