Skip to content

Between obligations and evidence: a participatory study translating the EU AI Act into technical requirements

Between Obligations and Evidence: A Participatory Study of Translating the EU AI Act into Technical Requirements

arXiv: 其他 AI 安全论文· arXiv:2611.18439· Helma Torkamaan, Masoud Mansoury, Yanan Xin, Oscar Oviedo Trespalacios, Roberta Savella, Andrej Zwitter, Maria Heuss, Huy Son Nguyen, Marie-Therese Sekwenz, Moyi Li, Roberta Tamponi, Megha KhoslaDelft University of Technology; Institute of Information Science and Technologies (ISTI), National Research Council (CNR); University of Pisa; University of Klagenfurt; University of Amsterdam; Leiden Institute of Advanced Computer Science (LIACS), Leiden UniversityIndexed by this site · 4 days ago· Originally publishedGovernance· Its place in the scan record →

Translation in progress — untranslated text is omitted.

Quick read

Other

Based on the paper PDF (AI-generated); refer to the original

AI summaryA participatory study organized workshops with 32 researchers, policymakers and practitioners from different disciplines, using real-world scenarios in healthcare, hiring and public transport to examine how the EU AI Act's obligations for high-risk AI systems are interpreted, translated into technical requirements and evaluated in practice. Thematic analysis identified several recurring tensions: fairness metrics may mask discrimination in hiring, time pressure and insufficient information may weaken human oversight in public transport, and changes in the patient population may invalidate previously established evidence of system performance. The authors argue that technical, organizational and evidentiary challenges are interdependent, and that establishing and maintaining compliance requires not only technical evaluation but also attention to the long-term use, governance and monitoring of AI systems; they discuss implications for supporting context-sensitive, continuous compliance.

Questions
The EU AI Act sets technical and organizational obligations for high-risk AI systems, but the provisions do not specify what should be measured, with which metrics, or what evidence counts as compliance; implementation requires contested interpretation in specific contexts.
Methods
The authors organized a full-day participatory workshop within SIG-AI ACT: 46 registrants submitted concerns beforehand, and 32 interdisciplinary participants completed worksheets, quizzes and fishbone-diagram adversarial analysis around three hypothetical high-risk scenarios—public transport, healthcare and recruitment—followed by reflexive thematic analysis to derive themes.
Experiments and results
The analysis yielded three types of structural tensions before the workshop and six themes after it: bias can come from data, design and deployment feedback; human-in-the-loop does not guarantee effective oversight; responsibility allocation and capacity to fulfill responsibilities are misaligned; environmental changes can invalidate existing compliance evidence; good technical metrics do not mean support for the intended decision; and systems may be repurposed without reassessment.
Limitations and future work
The paper does not have a Limitations section specifically discussing limitations. The study is based on a single-day workshop with 3 hypothetical scenarios and 32 participants; the conclusions come from qualitative thematic analysis and were not empirically validated on real deployed systems.

Related events and news

1 events · 1 news mentions

News that mentions this paper

研究提出智能体抓取网站合规基准

  • Luiza's Newsletter(AI 治理)cites the paper's id or link
    ScrapeBench:评测智能体抓取网站的法律合规
    5. “Between Obligations and Evidence: A Participatory Study of Translating the EU AI Act into Technical Requirements” by Helma Torkamaan et al. - link
Read originalarxiv.org