Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

A critical look at the EU AI Act’s requirements and affected systems : who must explain what?

Wagner, Matthias LU orcid ; Wittmann, Marie-Therese and Borg, Markus LU (2025) p.494-503
Abstract
The Artificial Intelligence (AI) Act is a relatively new regulation set out in the European Union (EU). It lays out various requirements for AI systems that are placed on the market in the EU. This paper provides a critical and differentiated view of its requirements from a software engineering perspective. A detailed analysis of the legal document shows what requirements are set out for different types of AI systems covered by the AI Act. This also includes how the risk classification of AI systems works, how this relates to the applicable requirements, and its interplay with the potential impact of this law. A comparative requirements analysis shows the variance in the extent of requirements coverage between the different types of AI... (More)
The Artificial Intelligence (AI) Act is a relatively new regulation set out in the European Union (EU). It lays out various requirements for AI systems that are placed on the market in the EU. This paper provides a critical and differentiated view of its requirements from a software engineering perspective. A detailed analysis of the legal document shows what requirements are set out for different types of AI systems covered by the AI Act. This also includes how the risk classification of AI systems works, how this relates to the applicable requirements, and its interplay with the potential impact of this law. A comparative requirements analysis shows the variance in the extent of requirements coverage between the different types of AI systems as defined by the AI Act, with a special focus on interpretability and explainability. We conclude that the AI Act marks a big leap in terms of general system transparency because of some transparency requirements with a wide-reaching scope of affected AI systems. At the same time, it is acknowledged that its most comprehensive and severe requirements, including the ones related to explainability and interpretability, only apply to a more narrow pre-defined set of AI systems. Upcoming harmonized standards are expected to clarify existing ambiguities and significantly influence the act’s impact and complexity of operationalization. (Less)
Please use this url to cite or link to this publication:
author
; and
organization
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
published
subject
host publication
2025 IEEE 33rd International Requirements Engineering Conference Workshops (REW)
pages
10 pages
publisher
IEEE
ISBN
979-8-3315-3835-4
979-8-3315-3834-7
DOI
10.1109/REW66121.2025.00074
language
English
LU publication?
yes
id
33d3aa08-acb8-446b-b1e5-62b9267283cd
date added to LUP
2025-10-16 15:52:37
date last changed
2025-10-24 03:34:14
@inproceedings{33d3aa08-acb8-446b-b1e5-62b9267283cd,
  abstract     = {{The Artificial Intelligence (AI) Act is a relatively new regulation set out in the European Union (EU). It lays out various requirements for AI systems that are placed on the market in the EU. This paper provides a critical and differentiated view of its requirements from a software engineering perspective. A detailed analysis of the legal document shows what requirements are set out for different types of AI systems covered by the AI Act. This also includes how the risk classification of AI systems works, how this relates to the applicable requirements, and its interplay with the potential impact of this law. A comparative requirements analysis shows the variance in the extent of requirements coverage between the different types of AI systems as defined by the AI Act, with a special focus on interpretability and explainability. We conclude that the AI Act marks a big leap in terms of general system transparency because of some transparency requirements with a wide-reaching scope of affected AI systems. At the same time, it is acknowledged that its most comprehensive and severe requirements, including the ones related to explainability and interpretability, only apply to a more narrow pre-defined set of AI systems. Upcoming harmonized standards are expected to clarify existing ambiguities and significantly influence the act’s impact and complexity of operationalization.}},
  author       = {{Wagner, Matthias and Wittmann, Marie-Therese and Borg, Markus}},
  booktitle    = {{2025 IEEE 33rd International Requirements Engineering Conference Workshops (REW)}},
  isbn         = {{979-8-3315-3835-4}},
  language     = {{eng}},
  month        = {{09}},
  pages        = {{494--503}},
  publisher    = {{IEEE}},
  title        = {{A critical look at the EU AI Act’s requirements and affected systems : who must explain what?}},
  url          = {{http://dx.doi.org/10.1109/REW66121.2025.00074}},
  doi          = {{10.1109/REW66121.2025.00074}},
  year         = {{2025}},
}