How to regulate the many uses of AI has been a growing concern of regulatory bodies. Fostering innovation while providing adequate protection against risks such as discrimination are the main challenge taken up by a regulatory attempts.

The presentation discusses the implications of the EU Proposal’s product-oriented regulatory design. It takes algorithmic credit scoring as a test case, drawing on prior comparative work with the US regulatory landscape. Advantages of the Proposal’s cautious attempt to focus on risk and quality management are highlighted. A deep dive into non-discrimination laws shows limits of this approach. It is submitted that the focus on seemingly formal-technical questions will not always allow to escape taking a stance on substantive values and norms.