LIST is strengthening its role as a European leader in the evaluation and governance of artificial intelligence.

The LIST AI Sandbox has established itself as the national reference platform for rigorously and transparently testing and evaluating AI systems. As part of the LLMs4EU project, a €40 million pan-European initiative coordinated by the Alliance for Language Technologies (ALT-EDIC) and involving 20 partner countries, a language assessment module dedicated to Luxembourgish has been directly integrated into the platform. This module tested the models using official exams from the Luxembourg National Institute of Languages (INLL), covering levels A1 to B2 of the Common European Framework. Of the 54 models evaluated, many were able to handle the elementary levels, but performance dropped sharply at levels B1 and B2, with only a handful of large models reaching the highest levels — often out of reach for European SMEs. This work, conducted in partnership with the University of Luxembourg and the Zenter fir d’Lëtzebuerger Sprooch (ZLS), demonstrated how technology can serve as a catalyst for linguistic diversity and cultural sovereignty.

Building on this expertise, LIST has begun developing, in collaboration with the University of Luxembourg’s SnT, an AI Assessment Sandbox Configurator: a modular open-source framework that simplifies the configuration and deployment of custom assessment environments, directly aligned with the requirements of the European AI Act. In October 2025, LIST contributed to the launch of the RE.M.I. (Regulation Meets Innovation) initiative, organized jointly by the National Commission for Data Protection (CNPD) and the Luxembourg AI Factory, which brought together more than 150 stakeholders from the Luxembourg ecosystem —including startups, large companies, public institutions, and regulators — to discuss AI Act compliance issues. It was in this context that the configurator was presented for the first time, targeting both AI providers wishing to conduct their own self-assessment, and competent authorities operating regulatory sandboxes.

These advances are accompanied by leading scientific recognition. In 2025, 110 articles on AI were published by LIST, representing 23% of its total scientific output. This momentum is also reflected in international awards. LIST researchers Alessio Buscemi and German Castignani received the Best Paper Award at the European Conference on Artificial Intelligence (ECAI 2025), a leading conference on artificial intelligence at the European level. Out of 2,672 submissions, only 626 papers were selected; theirs was among the three honoured with this prestigious award — a distinction that confirms the quality and scope of the research conducted at LIST in this rapidly evolving field.