autumn 2026 generator and tamper-proof evidence. It is intend- ed to be useful across the board: from public insti- tutions and large enterprises to startups and small and medium-sized companies that want to deploy AI responsibly. Why environment control matters Businesses handling sensitive information may not want to transfer assessment data to an external en- vironment they cannot fully control. They may also need to demonstrate where the data was processed, who accessed it and how results were recorded. As an open-source application, the AI Assess- ment Sandbox Configurator lets users keep more control over where assessment takes place and how evidence is managed. Organisations can as- sess AI on their own premises or in a private or sovereign cloud, with no vendor lock-in – a key distinction from conventional external testing ser- vices. It does not force every organisation into one hosting model. For international companies, this flexibility addresses a practical concern: meeting European expectations for trustworthy AI while retaining control over the testing environment. Built on experience The Configurator builds on research and testing experience developed by two of Luxembourg’s leading research organisations: the Luxembourg Institute of Science and Technology (LIST) and the University of Luxembourg’s Interdisciplinary Cen- tre for Security, Reliability and Trust (SnT). Both have extensive experience in the field. LIST has operated its own AI Sandbox – a hands-on envi- ronment in which organisations evaluate AI models for robustness, fairness, bias and regulatory compli- ance – for several years. Its users have included Ban- que Internationale à Luxembourg, the City of Luxem- bourg and Mistral AI. The SnT has worked in the field for a decade with industry and public-sector partners, including several local banks, the Luxembourg Minis- try of Digitalisation and European institutions. Their real-world experience has fed into an application that meets real user needs. Key features Although the AI Assessment Sandbox Configurator does not guarantee compliance or replace specialist advice or regulatory due diligence, it adds concrete value to a company developing new solutions for the European market. In particular, it offers: a technical environment to assess an existing AI system against requirements relevant to its use case; a structured method for assessing AI systems and generating evidence; a platform that facilitates multi-disciplinary collaboration; more control over where testing takes place; a connection to Luxembourg’s research and AI ecosystem. The Luxembourg AI Factory expects the tool to serve more broadly across Europe and to be further co-developed by researchers, regulators, certifica- tion bodies and companies in other countries. Sever- al pilots are already underway across the continent, and the underlying research has been published in peer-reviewed scientific papers. With applications in sectors such as healthcare, finance and manufactur- ing, it offers a practical model of AI sovereignty: par- ticipating in Europe’s AI market while retaining con- trol over data, infrastructure, testing and evidence. For companies looking to set up in Europe, the in- sight is clear: market access increasingly depends not only on what an AI system can do, but on whether its developer can demonstrate how it was tested and gov- erned. The AI Assessment Sandbox Configurator is de- signed to help make that demonstration possible. AI SOVEREIGNTY