TÜV奥地利：2025安全与可认证的AI：概念、挑战与经验总结白皮书（英文版）（63页）.pdf

1、arXiv:2509.08852v1 cs.CY 8 Sep 2025Safe and Certifiable AI Systems:Concepts,Challenges,and Lessons LearnedVienna,September 2025Kajetan Schweighofer4,Barbara Brune2,Lukas Gruber4Simon Schmid3,4Alexander Aufreiter3Andreas Gruber1Thomas Doms1,2Sebastian Eder2Florian Mayer2Xaver-Paul Stadlbauer2Christop

2、h Schwald2Werner Zellinger4Bernhard Nessler3Sepp Hochreiter41TRUSTIFAI GMBH2TV AUSTRIA HOLDING AG3Software Competence Center Hagenberg4Johannes Kepler University Linz-Institute for Machine LearningEqual ContributionImprint:TV AUSTRIA HOLDING AG,TV AUSTRIA-Platz 1,2345 Brunn am Gebirge,AustriaFigures

3、Figure 1 elenabsl|Shutterstock.AbstractThere is an increasing adoption of artificial intelligence in safety-critical applications,yet practical schemes for certifying that AI systems are safe,lawful and socially accept-able remain scarce.This white paper presents the TV AUSTRIA Trusted AI framework

4、an end-to-endaudit catalog and methodology for assessing and certifying machine learning systems.The audit catalog has been in continuous development since 2019 in an ongoing col-laborationwiththeInstituteforMachineLearningatJohannesKeplerUniversityLinz,which was further extended with the Software C

5、ompetence Center Hagenberg and theresulting joint-venture TRUSTIFAI.Building on three pillars Secure Software Development,Functional Requirements,and Ethics&Data Privacy the catalog translates the high-level obligations of the EUAI Act into specific,testable criteria.Its core concept of functional t

6、rustworthinesscouples a statistically defined application domain with risk-based minimum perfor-mance requirements and statistical testing on independently sampled data,providingtransparentandreproducibleevidenceofmodelqualityinreal-worldsettings.Wepro-vide an overview of the functional requirements