AI holds great transformative potential for virtually all businesses, and the financial sector is undoubtedly one of them. Although AI brings advantages, it can also amplify risks or pose new ones.
Focusing specifically on creditworthiness assessments and credit scoring, AI will enable more precise credit scoring systems. But risks are non-negligible: (1) the principle of interpretability is key in any AI system and, in some cases, AI models can be particularly opaque and function as a ‘black box’; and (2) as with any other system, AI models, if not properly trained with the adequate data, can perpetuate or amplify historical discrimination patterns. AI systems rely on an enormous quantity of data and if said data is incomplete or inaccurate, the AI systems’ outputs can be severely biased.
The EU responded with the introduction of the AI Act, which entered into force on 1 August 2024. However, its uncertainties could inhibit the regulation’s full potential. The Act’s risk-based approach, coupled with overlapping horizontal and vertical legislation, poses complex compliance challenges, particularly for credit assessments and scoring.
About the author:
Judith Arnal is Board member at the Bank of Spain, as well as a member of its Audit Committee. She is an Associate Senior Research Fellow at CEPS and ECRI, a Senior Research Fellow at the Elcano Royal Institute and a member of its Scientific Committee.